Repository navigation
deps: Bump the dev-tooling group across 1 directory with 28 updates - #18
Closed
dependabot[bot] wants to merge 148 commits into
Closed
dependabot[bot] wants to merge 148 commits into
dependabot[bot] wants to merge 148 commits into
Conversation
An adaptive agentic layer for Claude Code: right-sized orchestration, SDLC specialists, and durable memory that compounds. - 9 internal design docs (.docs/internal/01–09) - Roster: 8 agents + 13 skills (incl. conducting) + references - Conductor front door (/agentry) + 10 node commands - Plugin assembly: .claude-plugin manifests, .mcp.json (mem), SessionStart primer hook - pnpm monorepo: @agentry/core (typed contract) + @agentry/memory (scaffold) - check-plugin gate, biome, tsconfig base, MIT license Memory MCP implementation + benchmark harness are next; Workbench is V2. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- .docs/internal/10-execution-roadmap.md: phased, acceptance-gated build plan (0→completion) - CLAUDE.md: enforce the code bar (SRP, types-in-core, ports-and-adapters, right-sized) + standing rules - @agentry/core: split index.ts into enums/memory/artifacts/events/config; index is a pure barrel - plugin.json: add homepage + repository (github.com/Codestz/agentry) - pnpm install: lockfile committed (node_modules gitignored) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Clean ports-and-adapters (domain → application → persistence → resolution → tools): - domain: id (ULID + origin-qualified), ports, ranking, dedup - resolution: two-root (global ~/.agentry + project) + scope→origin - persistence: JSON file-store (source of truth) + in-memory node:sqlite/FTS5 index (rebuilt at startup — simpler than temp+swap; files remain truth) - application: MemoryService (write+dedup-reinforce, recall+prime, search, update, feedback usefulness-signal, stats, episode_write) + flows (distill/consolidate, propose-only) - tools: 9 thin MCP adapters + stdio server entry - 5 node:test units green + headless stdio smoke (server lists all 9 tools) - esbuild → committed dist/index.js (zero-install); check-plugin fully green Roadmap (doc 10): Phase 1 done headless; live /mcp reload check pending. Phase 2 next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude Code spawns Node 22, whose bundled SQLite lacks FTS5 → the `CREATE VIRTUAL TABLE … USING fts5` threw "no such module: fts5" → the mem server exited on startup (MCP -32000). Capability-first storage: try FTS5, fall back to a plain table + token scan when absent. Verified write→recall round-trips on both node 24 (FTS5) and node 22 (fallback). esbuild target → node22. dist rebuilt (lockstep). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Root .mcp.json double-loaded — once as the plugin config (CLAUDE_PLUGIN_ROOT defined, works) and once as PROJECT config when working inside the agentry repo (CLAUDE_PLUGIN_ROOT undefined → "Missing environment variables" → -32000 in /mcp). A dogfooding-only collision. Inlining mcpServers into plugin.json removes the standalone file the project loader chokes on; the plugin server is unaffected. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Honors doc 07's design (frontmatter + prose body) over the JSON I'd shipped: - file-store: <slug>-<ulid>.md, YAML frontmatter + body (text/task), via bundled `yaml` - one file per id (stale-slug sibling removed on rewrite); dedup-reinforce still prevents same-concept duplicates at write time — the ulid is identity, not a dup source - id: add slug(); index.ts + tests use MarkdownFileStore; 5/5 tests green; dist rebuilt - doc 07: document the on-disk format + the in-memory-index implementation note Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
yaml@2 is CommonJS; bundled into ESM its internal require("process") hit
esbuild's "Dynamic require not supported" → mem crashed on startup. Add a
--banner createRequire shim so esbuild's __require resolves a real require.
Verified write→recall on node 22 (the runtime Claude Code spawns) + 5/5 tests.
dist rebuilt (lockstep).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The inline esbuild command crossed the smell line once it needed a shell-escaped banner inside a JSON script. Move to packages/memory/build.mjs (esbuild JS API): readable, no escaping, room to grow (sourcemaps/minify/externals). package.json build → `node build.mjs`. dist rebuilt — banner intact, boots on node 22, 5/5 tests. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
doc 02 §3 promised manual removal but doc 07's tool surface never specced it, so pruning required a filesystem rm. Add memory_forget(id): deletes a memory from the live store and disk by id. FileStore.deleteFact/deleteEpisode ports + service.forget + tool + test (6/6 green). Updated doc 07 to 10 tools. dist rebuilt (lockstep). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Memory core done + live-verified (/mcp green, round-trip, 6/6, 10 tools). Phase 2 (dogfood) next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Prints the plugin.json version and exits 0 (handy for CI/release). First task built through /agentry (one-shot route — no orchestration tax). Gate unaffected. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dogfooding surfaced the conductor skipping designed gates: it inlined acceptance criteria (no spec.md) and dispatched the architect to produce plan+ADR+tasks in one pass — collapsing the spec gate and plan gate. Root cause was loose skill content, not the architect (which did its job). Fixes: - conducting: Spec is ALWAYS a written artifact; spec gate + plan gate are mandatory ordered stops; plan first → gate → then split (never bundled). - architect: when dispatched to plan, produce Plan + ADR only; split is a separate dispatch after the plan gate. Behavior lives in the skills, not operator memory — so the fix is in the skill. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bare /agentry isn't possible for a multi-skill plugin (plugins.md:45 — single-skill only). Rename the front-door command (commands/agentry.md → go.md) to a clean verb; update conducting skill description, the SessionStart primer message, README, CLAUDE.md. Namespaced nodes (/agentry:spec etc.) unchanged. Internal design docs still say /agentry conceptually (sweep later). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Onboard (explorer) caught two drifts via recall of the inline-MCP gotcha: - layout showed a root .mcp.json → it's inline in plugin.json now - "design docs (01–09)" → 01–10 Comprehension → finding → fix; the moat's recall drove the catch. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Grouped pass-checklist with per-section counts + FAIL/OK tally; ANSI color gated on TTY and NO_COLOR so piped/CI output stays plain. Built through /agentry:go (routed one-shot) and distilled into a repo memory. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Built through /agentry:go (decompose+verify: implementer ⇄ verifier). - core: Fact gains `decay` (strike counter) + `archived` status + `archivedAt`. - feedback(): recalled-but-unused decays usefulness + adds a strike; cite+pass resets strikes; archives at threshold (4, doc 02 §7 knob) only once usefulness floors at 0; suspect/fail path doesn't strike toward archive. - archived excluded from recall (task+prime), search, activeFacts; stats gains an `archived` bucket. recover() + memory_recover tool (11th) undo a false-archive and re-index. db-index.add() made idempotent so recover can't dup FTS rows (verifier-caught). 13 tests green; dist rebuilt (lockstep). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
From the decay+archive run — three gaps, all fixed in the durable layer: 1. Spec/plan/tasks must be PERSISTED as files in .agentry/work/<id>/ before gating — reasoning them in chat without the artifact is the failure. 2. Build loop rebuilds dist on packages/memory/src changes (dist-lockstep) — the live MCP runs the bundle, not src; check-plugin before the ship gate. 3. Fix-loop = re-dispatch a FRESH implementer with the verifier's fix contract, not SendMessage (which may be unavailable) — clean context, exact refs. conducting: gate-writes-file + Build-loop section + 2 anti-patterns. implementer: dist-lockstep edge case. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
From the decay+archive run, stated project-AGNOSTICALLY (Agentry ships into any repo; project specifics live in that repo's CLAUDE.md/memory, discovered): 1. Gates persist their artifact as a FILE in .agentry/work/<id>/ before gating (spec/plan/tasks) — not just reasoned in chat. 2. Build loop: keep build artifacts in sync with source + run the project's build/lint/tests/gate before ship — DISCOVER those commands from the repo, never hardcode them here. 3. Fix-loop = re-dispatch a FRESH implementer with the verifier's fix contract (don't depend on SendMessage) — clean context, exact refs. conducting: gate-writes-file, Build-loop section, 2 anti-patterns. implementer: generic "rebuild the artifact your source generates" edge case. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Folds in the parity insight: same model → average-task quality is parity, so measuring it proves nothing. Reframe around the four edges + edge-revealing metrics: - §1a "parity trap": escaped-defect rate (R1′ bug-seeded), requirement coverage (R2), warm−cold delta (R3), no-tax (R0). The moat is the only CATEGORICAL edge (plain can't recall) — land it first; the rest are rates over N runs. - §3 metrics: add escaped-defect + warm−cold; demote average AC-pass to substrate. - §4 C1–C4 sharpened to the edge metrics. §8: add R1′ bug-seeded regime. - Build the harness LAST (after dogfooding stabilizes the conductor/skills); interim signal = a manual duel. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…dump Per maintainer call — don't auto-inject the warm set at SessionStart (no task yet → generic top-N is low-relevance and taxes every session). The primer is a minimal nudge (counts + front-door + optional continue-context pointer, hard-capped); relevant memory enters context via the conductor's task-specific recall. Reframed doc 02 §2 (recall gate) + §5 (continue-context) + doc 07 §3 (priming hook). The primer code was already minimal — this aligns the docs to it and to the principle. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Setup, build/test/check commands, the code bar + process rules (dist-lockstep, reload-gated, run check before commit), commit convention. Built through /agentry:go (one-shot); project specifics recalled from CLAUDE.md/repo-facts + verified accurate against package.json and the tree. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
) * Actionable, consistent error handling across the 11 memory MCP tools Every @agentry/memory tool now returns one structured error envelope on a genuine failure, marked as an MCP error (isError:true) instead of a silent success flag — so a calling model that can only see the response can self-correct. What changed: - @agentry/core: new MemoryError contract (zod) — { code, what, why, fix } over a closed code set (bad-input | not-found | invalid-state | internal | storage-read). Single source of truth, barrel-exported. - memory/tools: err() serializer (sibling of ok()), a builder module (notFound/badInput/invalidState/storageRead/internal) as the one home for codes + prose, and shared adapter plumbing (guard/partial) in adapter.ts. - memory-service: update/forget/recover/feedback/distill now return typed discriminated outcomes; adapters map them to ok/err. recover splits not-found vs invalid-state. Semantic bad-input (blank id) is an error. - Partial tools (feedback, distill): all ids invalid -> error; some valid -> success reporting applied + skipped[{id,reason}]. - write surfaces a no-op supersedes as additive WriteResult.supersededMissing. - Persistence read failures are surfaced (ports ReadResult channel; file-store stops swallowing corrupt records) and reported via memory_stats.readErrors. Consistency is structural: one core type, one builder module, one serializer — nothing per-tool to drift. internal() takes no error argument, so a stack/path leak is structurally impossible. Tests: 33 memory + 5 core pass. dist rebuilt in lockstep; check-plugin clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Harden planning craft from the error-handling dogfood run Generic improvements surfaced while running the MCP error-handling task through /agentry:go (kept project-agnostic — shipped payload): - Task contracts gain `excludes` (must-not-touch), the explicit negative of `owns` — enforces the boundary from both sides. - Pin exact names on any surface a sibling consumes (the consumer codes against a frozen signature, not a paraphrase). - New anti-pattern "Unverified-state assumption" — a task must verify repo state and adjust, never assert an unchecked "absent"/"present" claim. - Tasks carry their leave-it-green verify/build steps, discovered from the repo (never hardcoded — the task names specifics, the skill stays generic). - Architect emits the structured task frontmatter explicitly (don't bury owns/exposes/deps/satisfies in prose — keep coverage machine-checkable). - Canonical work-folder layout: ADRs as a folder (adr/NNN-slug.md, a numbered append-only series), mirroring tasks/. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The partial() all-invalid path joined skipped ids and fed them to the single-id notFound() builder, so the `why` read `id "g:A, g:B"` — multiple ids looking like one. Add a notFoundMany() builder (same code:"not-found" and field set, AC4) that lists each id distinctly quoted; partial() uses it for >1 id and keeps notFound() for the single-id case. Strengthened the all-invalid test to assert each id appears quoted (regression guard). Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The dist-lockstep check compared dist-vs-src mtimes, but git doesn't preserve mtimes — a clone, checkout, or squash-merge rewrites them, producing a false "dist is older than src" warning when the bundle was never actually stale (seen twice after merging PRs #1 and #2). Repeated false warnings train contributors to ignore the gate. Fix: build.mjs stamps a SHA-256 of all src file contents into dist/.srchash; check-plugin recomputes and compares. Checkout-proof, and it catches the real failure (src edited but dist not rebuilt). Shared scripts/lib/src-hash.mjs so the two sides can't drift. Verified: touching a src file (mtime only) stays "up to date"; a content change without rebuild reports "stale". Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dogfood finding: a fresh-session conductor one-shotted "emit events to events.jsonl" — skipping the spec/plan gates — and silently guessed the crux design choice (resolved a hook's work-id by "newest-mtime folder", a fragile heuristic). Root cause: the signals→shape rubric is a flat, equal-weight matrix. 4/5 signals said "floor" (few files, reversible, precedent exists), only Unknowns said "escalate" — so it averaged to one-shot. Bias-to-floor was one-directional (a vivid over-routing failure, no under-routing counterweight), and escalation triggers were mid-flight-only. Fix (generic, durable layer): - Override rule: the HIGHEST-severity signal sets the floor, not the average. A real unknown or an undecided design fork VETOES one-shot → >= spec-first regardless of file count. Small footprint != small decision. - New signal "Decision-content" (is there an undecided fork?), decoupled from Scope, so a 2-file task with a load-bearing choice routes correctly. - Pre-flight fork check before any edit (triggers are no longer mid-flight-only). - Symmetric framing in go.md: pair the one-liner over-routing failure with the under-routing one (cite the mtime work-id incident). - Broaden the spec-gate trigger from "under-specified" to "under-specified OR specified-goal-with-undecided-mechanism". Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…or/model now apply) All 8 agents had a plain multi-line `description:` containing <example> blocks. That's malformed YAML — it breaks at the first inline colon (e.g. "Examples:", "user:"), so every frontmatter field AFTER the description (model, color, skills) was silently dropped at load. Net effect: agents were NOT preloading their craft skills, NOT applying their color, NOT honoring model — we'd been running the whole roster on the body prompt alone. Fix (proven against the official plugin-dev/agent-creator.md reference): - `description: |` block scalar (2-space indent) so example-laden text is literal and can't break parsing. - Clean key:values first (name/model/color/skills), description last. - Distinct, valid colors per agent (magenta isn't a valid value; architect +product-owner and explorer+researcher had collided): architect=blue designer=purple explorer=cyan implementer=green librarian=yellow product-owner=orange researcher=pink verifier=red Verified: all 8 frontmatters parse with name/model/color/skills/description; check-plugin 6/0. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Capability matrix for plugin subagents (what's usable vs ignored), the block-scalar frontmatter standard (the fix landed in ec5a2e8), the corrected Workbench events mechanism (plugin-level SubagentStart/SubagentStop, not per-agent hooks), deeper references per skill, and the roster fragility audit. Draft — pending sign-off. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The fix-loop guidance hedged "SendMessage may be unavailable in this environment" — a conditional reference to a tool that's gated behind an experimental flag and absent in most harnesses. Stale, confusing guidance. Replace with the durable principle on its own merits: a fresh implementer carrying the verifier's fix contract is reliable + reproducible; don't rely on resuming a prior agent's muddied context. No tool name, fully generic. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The architect dropped its ADR at the work-dir root: the adr/-folder convention
lived only in the conductor's command/skill, not in the architect's own
instructions — so when dispatched, its craft never carried it. The conducting
layout line was also ambiguous ("single docs stay at the root" → one ADR read
as a single doc).
Fix: architect's output contract now names the exact path
(.agentry/work/<id>/adr/NNN-slug.md, always a folder); conducting layout line
disambiguated — adr/ and tasks/ are always folders (numbered append-only
series), even with one entry; spec.md/plan.md stay at root.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Plugin-level hooks emit a WorkEvent per subagent start/stop into .agentry/work/<id>/events.jsonl, work-id resolved by a session-keyed pointer the PostToolUse binder writes (ADR-001 — no mtime, no marker). - core: WorkEvent gains additive agentId?/session? (correlate start↔stop, partition by session) - hooks/work-id-binder.mjs: PostToolUse Write|Edit → session_id→workId pointer at .agentry/run/sessions/<sid>.json (atomic tmp+rename) - hooks/subagent-emit.mjs: SubagentStart/Stop → resolve pointer, append one JSONL WorkEvent; unresolved → silent skip - hooks.json: events nested under top-level "hooks" wrapper (the bare form silently fails to load — even SessionStart never fires); commands quoted - dep-free node builtins, always exit 0 (non-blocking), silent in non-Agentry repos Live-fired after relaunch: AC1 binder→pointer, AC2/AC4 Subagent*→events, agentId correlation, AC3 one-object-per-line. 21 hook + 3 core tests green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The channels docs previously presented only --dangerously-load-development-channels as the way to enable channels. That flag bypasses Anthropic's approved allowlist and is for custom / in-development channels — Agentry's own bridge needs it today because it isn't officially listed. Official allowlisted channels (Telegram/Discord/iMessage) use --channels plugin:<name>@<marketplace>. Document both paths across the install, channels, and troubleshooting pages. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…er fix, run-guard hook, flow-compliance eval After v2.0 (Flow MCP, Workbench, channels) the harness described its process but didn't enforce it — proven live when the conductor bypassed Flow + spec-first on a multi-file build. This wires the whole surface to Flow and adds teeth + measurement. A — wiring + the frontmatter fix: - go.md mandates run_start + event_emit(routing-decision) as the first above-floor action; the 7 node commands reference their real Flow tools. - architect.md + planning skill author task bodies as plain markdown (## Contract, no leading ---); fixes the double-frontmatter that rendered broken in the Workbench. - 5 specialist agents gain role-scoped Flow I/O + a Workbench/channel steering note; reviewing/integrating/shipping skills learn the review loop + permission relay. B — teeth: - plugin/hooks/flow-run-guard.mjs (dep-free PostToolUse) flags a flow-skipped marker when a gate artifact is written with no run; respects the one-shot floor. C — measurement: - packages/eval flow-compliance probe asserts on the events/artifact surface (run_start before dispatch, spec before tasks, single frontmatter, no skip flag); above-floor detection is artifact-based. Caught a real regression on its first run. - packages/flow task-file-store defensively strips a leading --- block; flow + workbench-server dists rebuilt (dist-lockstep). Live-verified after restart: the hook flags a bypass, a correctly-conducted run PASSes the probe (single-frontmatter proven on a deliberately-malformed task body), and a run that skips the routing event correctly FAILs. check-plugin 10/10; flow 132, eval 191, hook 18. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…gitignored .agentry/work tree Three workbench-server tests (fs-work-repository, event-store REAL layer, work-reader) read the live .agentry/work/<run>/ tree as their 'real shapes' fixture. That tree is gitignored (.gitignore: .agentry), so a clean CI checkout has no runs and the tests fail the 'fixture run dir present / at least one run present' preconditions — red CI on every push, including the v0.2.0 tag. Commit the reference run as tracked fixture data under test/fixtures/work-fixture/ (no .agentry segment, so not ignored) + a loader that copies it into a temp .agentry/work/ at setup; repoint the three REAL-layer tests at the temp cwd. Content unaltered, so the asserted shapes are preserved. Workbench-server 123/123 with the live tree hidden. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…, docs site, harness enforcement, CI fix) The root CHANGELOG.md stopped at 0.1.2 and the web changelog page predated the harness work. Add the full 0.2.0 entry (Added/Changed/Fixed) to both, and the compare links. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…, results-gated) Restructure @agentry/eval from 5 accreted probes into 3 public probes — Moat (warm-vs-cold, the categorical headline), Right-sizing (results-gated routing), and Honesty (overclaim + flow-compliance) — plus a shared judge engine and relocated conduct primitives. Retire the bare/raw-Claude comparison: the bench is Agentry-value-only (routing precision + the moat + cost), not "beats bare Claude". - Unify routing + outcome into ONE conduct-and-judge harness: conduct once, capture the routed shape AND judge the produced result, score results-gated (lighter+good=win, lighter+bad=under-route-failure, heavier+good=tax, no-signal=indeterminate). Kills the bare label-match accuracy bug. - Extract one shared judge engine (judge/); relocate shared conduct primitives (conduct/); park quality/; delete superseded routing/outcome/outcome-judge/ flow-compliance. - De-bare the matrix; derive the cell/run label from --model (no hardcoded Sonnet). - Pre-register thresholds X (right-sizing) + W (moat) as falsifiable forms (calibration-pending) — never invented numbers. - Re-lead the report IA: Moat -> Right-sizing -> Honesty; drop the bare delta, the routing label-match, and the parked decision-quality from the public page. - Author 8 dual fixtures (defended floor + golden/broken/oracle). - Harden the judge parser to tolerate fenced/prose-wrapped JSON; sharpen 4 fixture brokens so the discrimination controls clear with margin. Built via Agentry's own process (spec -> plan -> split -> 10 tasks -> assemble-verify). R1 conduct-once validated live (rs-clamp -> one-shot, judged 1.00). First full calibration baseline ran clean but SATURATED (8/8 one-shot, all 1.00); a validity probe confirmed routing capture works headless (a hard task escalated to spec-first), so the fixtures are too easy, not the instrument broken — harder discriminating fixtures are the next step before any public run. 413/413 tests green, typecheck clean, check-plugin OK. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…xture A 3-module in-memory job queue (store + retry/dead-letter worker + index) with a defended spec-first floor. Live-verified: the conductor escalates it to spec-first (not one-shot), judged 1.00 → right-sizing-success, and the no-retry broken discriminates cleanly from golden at the controls gate. Adds the missing non-one-shot tier the saturated baseline lacked (the 8 original fixtures all one-shot). seed fails / golden passes / broken fails the oracle; 416/416 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A 5-module in-memory relational store (schema/table/query/relations/index with cross-cutting referential integrity). Defended floor: decompose. Live finding: the conductor routed it spec-first (not decompose) and built it perfectly (judged 1.00 → right-sizing-success) — evidence the model's decompose threshold is high and it does NOT over-orchestrate even substantial multi-module systems. Controls discriminate (broken fails 4 oracle tests on 3 planted defects); seed fails / golden passes / broken fails; 419/419 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…roadcast) Workbench review-comment (and human status-change) channel notifications were emitted UNCONDITIONALLY for every run's sidecar — and since each session's stdio flow process watches the same `.agentry/work/*` tree, a comment on run A spammed every other session, including ones conducting unrelated runs. Now session-targeted, with a safe orphan-broadcast fallback so no comment is lost: - process-identity.ts: per-process holder of the runs/session this process has operated on (a stdio MCP process is 1:1 with a session, so process-global identity == session identity). Captured at the top of resolveRunContext from the explicit `run` (ADR-005 normal path) + optional `session_id` args. - comment-routing.ts: pure shouldRouteToThisSession — route if the run is ours (myRuns or our persisted session pointer); drop if claimed by another session; broadcast if orphan (no live match, no pointer owner) so a Workbench-only run's comment is never lost. - run-pointer.ts: sessionsBoundTo(cwd, run) — which sessions the binder bound to a run (the "is this run claimed by anyone" signal). - ChannelBridge + StatusBridge take an optional shouldEmit gate (default always-on, preserving every existing test); applied after the seen/consume step so a not-ours comment is consumed once and never re-evaluated. 150/150 flow tests green, typecheck clean. Rebuilt plugin/flow/index.js (+ .srchash) and refreshed plugin/workbench/server/.srchash (dist-lockstep); check-plugin 0/0. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… expanded conductor prompt runWarm built `/agentry:go <task>` — the bare slash form, which is a 0-turn no-op under `claude -p` (the model never runs). So every moat warm conduct settled with num_turns:0, never routed, never recalled — the seed-did-not-land aborts were an artifact, and the moat was never validly measured. Now uses buildConductorPrompt (the expanded go.md, same fix the right-sizing probe already had) + the moat-specific memory-priming directive. Verified: the conductor now runs (5 turns). NOTE: a SECOND blocker remains (separate fix) — the conduct loads --plugin-dir but NOT the plugin's mem/flow MCP servers, so memory_recall is unavailable; seed-landing still fails until the mem MCP is loaded into the conduct. 419/419 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…(discrimination 1.0) `--plugin-dir` loads the plugin's skills/commands/agents but NOT its MCP servers, so every conduct ran with NO memory faculty — the conductor had no memory_recall tool, so the moat seed could never land (the mcp_servers at init showed only the host's globals, no agentry_mem). Added Invocation.mcpConfig → `claude -p --mcp-config`, and the moat probe now passes the bundled mem server (plugin/mem/index.js), which roots project memory at the conduct cwd (the sandbox workingDir where seedFact wrote). Result, with both this + the 0-turn fix: agentry_mem loads, memory_recall fires (6×), seed lands 2/2, and the moat SCORES — compoundRate 1.0, decoyLightenRate 0, discrimination 1.0: the relevant fact collapses the fork (warm→one-shot) while the irrelevant decoy holds (→spec-first). The categorical edge, controls-gated. N=2, directional. 419/419 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ariance conductor The moat effect is real but per-task conductor routing is stochastic, so a single k=1 draw swings wildly (discrimination 1.0 -> abort across runs). Now reports RATES over k repeats: compoundRate = (relevant repeats that LANDED and routed lighter) / (relevant repeats that LANDED) — a non-landed relevant repeat is indeterminate (recall never fired) and excluded; seedLandingRate is the validity readout; decoyLightenRate is the false-positive base rate; discrimination = compoundRate - decoyLightenRate, with per-task spread. Seed-landing gate is now a RATE FLOOR (>=0.5) not all-or-nothing. CLI: run moat --runs <k> --model <id>. Also: mandatory-recall directive + 6 project-specific forks (industry-default forks the model already knows leaked the decoy; swapped for genuinely-unguessable project decisions). 425/425 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…RKING code, not just a lighter route Closed a measurement hole: moat `compounded` was route-only (landed && lighter), so memory making the conductor skip the spec into a BROKEN one-shot would still score as a moat win. Now `compounded = landed && lighter && resultGood` — the warm-relevant one-shot's produced tree is judged (shared outcome rubric + an apply-the-decision clause that feeds the recalled fact to the judge), and only a lighter route that ALSO produces correct code counts. A landed+lighter+BAD run is surfaced as a distinct `lighterButBad` signal (never silently a win). Mirrors rightsizing's results-gating. CLI: --judge-model. 428/428 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The moat conducts are independent (fresh sandbox each) but ran strictly serial, so a k=3 x 6-task run was 36 serial conducts (~60-90 min). Now flattens the (task x repeat x arm) conducts into work items with pre-assigned unique idSeeds and drives them through mapPool(concurrency) — mirroring the rightsizing probe. Results reassemble deterministically (a staggered-runner test asserts parallel outcomes == serial, and all idSeeds unique). CLI: run moat --concurrency <n> (default 1; keep 3-4 on live Opus runs). Orchestration-only — gates/aggregation/result-quality untouched, same numbers. 431/431 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ecall tool_use Seed-landing aborted 0/18 on Opus even though the seeded fact demonstrably reached the conductor: the conductor accessed it via the Read tool (reading the seeded .agentry/memory/facts/*.md in its working dir) instead of the memory_recall MCP tool, and the detector required a memory_recall tool_use to register recall. Now a relevant run LANDS when its distinctive signature SURFACES in the run (recall MCP OR file-read of the project-memory store — both valid durable-memory access; the decoy control attributes the routing effect to the relevant fact's CONTENT, not the access mechanism). Requiring the tool_use was too narrow. 431/431 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ench report Pre-registered targets calibrated from the first measured runs: moat W=0.5 (moat-n6 discrimination 0.67, N=6, seed 6/6, gated — clears it; moat-mem 1.0 N=2 confirms); rightsizing X=0.8 (rs-baseline-k3 right-sizing-success 1.0, over-route-tax=0, under-route-failure=0 — clears it, with the saturation caveat carried in the note). Generated the self-contained dashboard results/2026-06-23/rs-baseline-k3/index.html stitching all three probes (moat headline 0.67, right-sizing no-tax, honesty 0/30 overclaim). Zero-API. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…roxy) The moat measured routing-shape-lightening (memory makes the conductor route lighter) — a weak proxy for memory's real value (error-prevention / better decisions), as the reshape concluded. Deleted src/moat/ + all wiring: the `run moat` CLI subcommand, report/moat.ts, MoatData, the loadMoat call, the moat template section. Kept the back-compat "moat" RunKind literal so historical run dirs still read. First step of reshaping the eval into a value-axis quality bench (plan: ok-so-now-what). 393/393 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…lity axes) src/bench/: conduct Agentry once per fixture, then score four ABSOLUTE axes from that single run (no baseline, no routing label-match): A decisionQuality — judge the decision trail (.agentry/work spec/plan/adr) with the 5-dim rubric B codeQuality + correctnessPassRate — judge the produced tree (4-dim rubric) + the hidden oracle C overclaimRate — said-done but judged-bad or oracle-fail (trust it unreviewed) D escapedDefectRate + verifyFireRate — on bug-prone fixtures, a bug shipped past the verifier Controls-first (code golden↔broken + planted decision gold↔poor) gate the judges before any score. Pure scorer (bench/score.ts), conduct-once probe (bench/probe.ts), rubrics re-exported from the existing engine. Zero-API tests (fake runner + canned judge over a bench-mini fixture). 416/416 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Phase 5: `run bench` subcommand drives runBenchProbe (RunKind += "bench"), mirroring run rightsizing — --fixtures-dir/--fixture/--k/--concurrency/--model/ --judge-model/--plugin-dir, decisionControlsDir = <fixtures>/_controls. run rightsizing kept (removed in the cleanup step). Phase 3: 10 realistic, harder fixtures under fixtures/bench/ — 5 bug-prone (float- money, CSV-quoting, business-days off-by-one, LRU-vs-FIFO, tier-boundary), 5 with a genuine hidden decision (rate-limiter window, retry policy, due-date semantics, LRU use-on-get), 3 multi-file with real seams, 1 trivial. Each: seed FAILS / golden PASSES / broken FAILS its hidden oracle (bug-prone oracles carry a subtle-bug probe). Shared Axis-A decision gold/poor controls. Loader test asserts the corpus + spread. 455/455 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Rebuilt the report (the old one was ugly) around the 4-axis bench. PageData now leads with BenchData; src/report/load-bench.ts narrows the latest scored kind:"bench" run; the new template.html is two tiers — a persuasive public SUMMARY (4-axis scorecard with plain-English claims + mean±std·n + the showcase strip of descriptive value: structure/memory/specialists/Workbench) over a technical DRILL-DOWN (per-task census, axis distribution, controls-passed block). Designer-verified in a browser (both tiers screenshot-checked); WCAG AA contrast (bumped --faint, a neutral, to clear AA). Rewrote docs/evals prose to the 4 axes. rightsizing/honesty subsumed (loaders kept for the cleanup step). 452/452 green, astro check clean, token-drift OK. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…robes (bench subsumes them)
The bench harness now owns all four axes, so the old probes are dead. Moved the shared
conduct primitives oracle.ts + compliance.ts to src/conduct/ (bench imports them);
deleted src/rightsizing/, src/report/{rightsizing,honesty}.ts, the honesty/quality
probe sources (kept quality/rubric.ts — bench's DECISION_RUBRIC), the run rightsizing
CLI handler + cellsFromFlag + --cell, the dead readRunInputs/fixture-routing pipeline,
and fixtures/rightsizing. Public CLI is now `run bench | replay | report`. Kept the
historical RunKind literals for back-compat reads. 327/327 green, check-plugin OK.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…5 harder fixtures The full bench surfaced two real things: (1) the Axis-B code judge can be fooled by clean-but-wrong code (gave bn-lru-cache 1.00 while the hidden oracle failed) — the meter working as intended; (2) the verifier never fired in any conduct, because verify is route-coupled (decompose-only) + prompt-described, not a mandated harness step. Measuring "did verify fire" graded a non-feature, so it's removed. - Strip verifyFireRate + verifyFired everywhere (score/probe/report-model/load-bench/ template/tests). The OUTCOME that mattered — a defect shipped under a "done" claim — survives as escaped-defect, folded into honesty (Axis C, the bug-prone slice of overclaim). - Repoint the Axis-B code-discrimination control from golden↔broken to golden↔SEED: a `broken` build is clean-but-functionally-wrong (judge rightly rates it MEDIUM on CODE QUALITY, so it can't clear the LOW bar) — that's the oracle's job; the seed (target unimplemented) is genuinely poor CODE. This is why the bench kept aborting at the trust-gate. - parseScore: recover the dimensions block when a long rationale truncates the JSON tail (the dominant real-judge flake); bound the rationale to one sentence. The grade is the dimensions; a mangled rationale tail no longer discards a valid verdict. - +5 fixtures (10 → 15): bn-job-queue, bn-event-ledger, bn-middleware-pipeline, bn-dag-scheduler (decompose-scale, 5-module, hidden cross-cutting decision + subtle integration bug) and bn-interval-merge (spec-first, boundary off-by-one). Each: seed fails / golden passes / broken fails its hidden oracle; all 5 discriminate gold 1.00 ↔ seed 0.00 with the real judge. 343/343 green, typecheck + check-plugin clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bumps the dev-tooling group with 28 updates in the / directory: | Package | From | To | | --- | --- | --- | | [@biomejs/biome](https://github.com/biomejs/biome/tree/HEAD/packages/@biomejs/biome) | `1.9.4` | `2.5.10` | | [tsx](https://github.com/privatenumber/tsx) | `4.22.4` | `4.23.12` | | [typescript](https://github.com/microsoft/TypeScript) | `5.9.3` | `7.0.2` | | [zod](https://github.com/colinhacks/zod) | `3.25.76` | `4.4.3` | | [@types/node](https://github.com/DefinitelyTyped/DefinitelyTyped/tree/HEAD/types/node) | `24.13.2` | `26.2.0` | | [@modelcontextprotocol/sdk](https://github.com/modelcontextprotocol/typescript-sdk) | `1.29.0` | `1.30.0` | | [chokidar](https://github.com/paulmillr/chokidar) | `4.0.3` | `5.0.0` | | [esbuild](https://github.com/evanw/esbuild) | `0.24.2` | `0.28.2` | | [astro](https://github.com/withastro/astro/tree/HEAD/packages/astro) | `6.4.8` | `7.2.4` | | [@astrojs/check](https://github.com/withastro/astro/tree/HEAD/packages/language-tools/astro-check) | `0.9.9` | `0.9.10` | | [ws](https://github.com/websockets/ws) | `8.21.0` | `8.21.3` | | [@codemirror/lang-markdown](https://github.com/codemirror/lang-markdown) | `6.5.0` | `6.5.2` | | [@dagrejs/dagre](https://github.com/dagrejs/dagre) | `1.1.8` | `3.1.1` | | [@tiptap/extension-link](https://github.com/ueberdosis/tiptap/tree/HEAD/packages/extension-link) | `2.27.2` | `3.30.3` | | [@tiptap/pm](https://github.com/ueberdosis/tiptap/tree/HEAD/packages/pm) | `2.27.2` | `3.30.3` | | [@tiptap/react](https://github.com/ueberdosis/tiptap/tree/HEAD/packages/react) | `2.27.2` | `3.30.3` | | [@tiptap/starter-kit](https://github.com/ueberdosis/tiptap/tree/HEAD/packages/starter-kit) | `2.27.2` | `3.30.3` | | [@uiw/react-codemirror](https://github.com/uiwjs/react-codemirror) | `4.25.10` | `4.25.11` | | [@xyflow/react](https://github.com/xyflow/xyflow/tree/HEAD/packages/react) | `12.11.0` | `12.11.3` | | [markdown-it](https://github.com/markdown-it/markdown-it) | `14.2.0` | `15.0.0` | | [react](https://github.com/react/react/tree/HEAD/packages/react) | `18.3.1` | `19.2.8` | | [@types/react](https://github.com/DefinitelyTyped/DefinitelyTyped/tree/HEAD/types/react) | `18.3.31` | `19.2.18` | | [react-dom](https://github.com/react/react/tree/HEAD/packages/react-dom) | `18.3.1` | `19.2.8` | | [@types/react-dom](https://github.com/DefinitelyTyped/DefinitelyTyped/tree/HEAD/types/react-dom) | `18.3.7` | `19.2.5` | | [react-router-dom](https://github.com/remix-run/react-router/tree/HEAD/packages/react-router-dom) | `6.30.4` | `7.18.2` | | [@types/markdown-it](https://github.com/DefinitelyTyped/DefinitelyTyped/tree/HEAD/types/markdown-it) | `14.1.2` | `14.2.0` | | [@vitejs/plugin-react](https://github.com/vitejs/vite-plugin-react/tree/HEAD/packages/plugin-react) | `6.0.2` | `6.1.0` | | [vite](https://github.com/vitejs/vite/tree/HEAD/packages/vite) | `8.0.16` | `8.2.2` | Updates `@biomejs/biome` from 1.9.4 to 2.5.10 - [Release notes](https://github.com/biomejs/biome/releases) - [Changelog](https://github.com/biomejs/biome/blob/main/packages/@biomejs/biome/CHANGELOG.md) - [Commits](https://github.com/biomejs/biome/commits/@biomejs/biome@2.5.10/packages/@biomejs/biome) Updates `tsx` from 4.22.4 to 4.23.12 - [Release notes](https://github.com/privatenumber/tsx/releases) - [Changelog](https://github.com/privatenumber/tsx/blob/master/release.config.cjs) - [Commits](privatenumber/tsx@v4.22.4...v4.23.12) Updates `typescript` from 5.9.3 to 7.0.2 - [Release notes](https://github.com/microsoft/TypeScript/releases) - [Commits](microsoft/TypeScript@v5.9.3...v7.0.2) Updates `zod` from 3.25.76 to 4.4.3 - [Release notes](https://github.com/colinhacks/zod/releases) - [Commits](colinhacks/zod@v3.25.76...v4.4.3) Updates `@types/node` from 24.13.2 to 26.2.0 - [Release notes](https://github.com/DefinitelyTyped/DefinitelyTyped/releases) - [Commits](https://github.com/DefinitelyTyped/DefinitelyTyped/commits/HEAD/types/node) Updates `@modelcontextprotocol/sdk` from 1.29.0 to 1.30.0 - [Release notes](https://github.com/modelcontextprotocol/typescript-sdk/releases) - [Commits](modelcontextprotocol/typescript-sdk@v1.29.0...1.30.0) Updates `chokidar` from 4.0.3 to 5.0.0 - [Release notes](https://github.com/paulmillr/chokidar/releases) - [Commits](paulmillr/chokidar@4.0.3...5.0.0) Updates `esbuild` from 0.24.2 to 0.28.2 - [Release notes](https://github.com/evanw/esbuild/releases) - [Changelog](https://github.com/evanw/esbuild/blob/main/CHANGELOG-2024.md) - [Commits](evanw/esbuild@v0.24.2...v0.28.2) Updates `astro` from 6.4.8 to 7.2.4 - [Release notes](https://github.com/withastro/astro/releases) - [Changelog](https://github.com/withastro/astro/blob/main/packages/astro/CHANGELOG.md) - [Commits](https://github.com/withastro/astro/commits/astro@7.2.4/packages/astro) Updates `@astrojs/check` from 0.9.9 to 0.9.10 - [Release notes](https://github.com/withastro/astro/releases) - [Changelog](https://github.com/withastro/astro/blob/main/packages/language-tools/astro-check/CHANGELOG.md) - [Commits](https://github.com/withastro/astro/commits/@astrojs/check@0.9.10/packages/language-tools/astro-check) Updates `ws` from 8.21.0 to 8.21.3 - [Release notes](https://github.com/websockets/ws/releases) - [Commits](websockets/ws@8.21.0...8.21.3) Updates `@codemirror/lang-markdown` from 6.5.0 to 6.5.2 - [Changelog](https://github.com/codemirror/lang-markdown/blob/main/CHANGELOG.md) - [Commits](https://github.com/codemirror/lang-markdown/commits) Updates `@dagrejs/dagre` from 1.1.8 to 3.1.1 - [Release notes](https://github.com/dagrejs/dagre/releases) - [Changelog](https://github.com/dagrejs/dagre/blob/master/changelog.md) - [Commits](dagrejs/dagre@v1.1.8...v3.1.1) Updates `@tiptap/extension-link` from 2.27.2 to 3.30.3 - [Release notes](https://github.com/ueberdosis/tiptap/releases) - [Changelog](https://github.com/ueberdosis/tiptap/blob/main/packages/extension-link/CHANGELOG.md) - [Commits](https://github.com/ueberdosis/tiptap/commits/v3.30.3/packages/extension-link) Updates `@tiptap/pm` from 2.27.2 to 3.30.3 - [Release notes](https://github.com/ueberdosis/tiptap/releases) - [Changelog](https://github.com/ueberdosis/tiptap/blob/main/packages/pm/CHANGELOG.md) - [Commits](https://github.com/ueberdosis/tiptap/commits/v3.30.3/packages/pm) Updates `@tiptap/react` from 2.27.2 to 3.30.3 - [Release notes](https://github.com/ueberdosis/tiptap/releases) - [Changelog](https://github.com/ueberdosis/tiptap/blob/main/packages/react/CHANGELOG.md) - [Commits](https://github.com/ueberdosis/tiptap/commits/v3.30.3/packages/react) Updates `@tiptap/starter-kit` from 2.27.2 to 3.30.3 - [Release notes](https://github.com/ueberdosis/tiptap/releases) - [Changelog](https://github.com/ueberdosis/tiptap/blob/main/packages/starter-kit/CHANGELOG.md) - [Commits](https://github.com/ueberdosis/tiptap/commits/v3.30.3/packages/starter-kit) Updates `@uiw/react-codemirror` from 4.25.10 to 4.25.11 - [Release notes](https://github.com/uiwjs/react-codemirror/releases) - [Commits](uiwjs/react-codemirror@v4.25.10...v4.25.11) Updates `@xyflow/react` from 12.11.0 to 12.11.3 - [Release notes](https://github.com/xyflow/xyflow/releases) - [Changelog](https://github.com/xyflow/xyflow/blob/main/packages/react/CHANGELOG.md) - [Commits](https://github.com/xyflow/xyflow/commits/@xyflow/react@12.11.3/packages/react) Updates `markdown-it` from 14.2.0 to 15.0.0 - [Changelog](https://github.com/markdown-it/markdown-it/blob/master/CHANGELOG.md) - [Commits](markdown-it/markdown-it@14.2.0...15.0.0) Updates `react` from 18.3.1 to 19.2.8 - [Release notes](https://github.com/react/react/releases) - [Changelog](https://github.com/react/react/blob/main/CHANGELOG.md) - [Commits](https://github.com/react/react/commits/v19.2.8/packages/react) Updates `@types/react` from 18.3.31 to 19.2.18 - [Release notes](https://github.com/DefinitelyTyped/DefinitelyTyped/releases) - [Commits](https://github.com/DefinitelyTyped/DefinitelyTyped/commits/HEAD/types/react) Updates `react-dom` from 18.3.1 to 19.2.8 - [Release notes](https://github.com/react/react/releases) - [Changelog](https://github.com/react/react/blob/main/CHANGELOG.md) - [Commits](https://github.com/react/react/commits/v19.2.8/packages/react-dom) Updates `@types/react-dom` from 18.3.7 to 19.2.5 - [Release notes](https://github.com/DefinitelyTyped/DefinitelyTyped/releases) - [Commits](https://github.com/DefinitelyTyped/DefinitelyTyped/commits/HEAD/types/react-dom) Updates `react-router-dom` from 6.30.4 to 7.18.2 - [Release notes](https://github.com/remix-run/react-router/releases) - [Changelog](https://github.com/remix-run/react-router/blob/react-router-dom@7.18.2/packages/react-router-dom/CHANGELOG.md) - [Commits](https://github.com/remix-run/react-router/commits/react-router-dom@7.18.2/packages/react-router-dom) Updates `@types/markdown-it` from 14.1.2 to 14.2.0 - [Release notes](https://github.com/DefinitelyTyped/DefinitelyTyped/releases) - [Commits](https://github.com/DefinitelyTyped/DefinitelyTyped/commits/HEAD/types/markdown-it) Updates `@types/react` from 18.3.31 to 19.2.18 - [Release notes](https://github.com/DefinitelyTyped/DefinitelyTyped/releases) - [Commits](https://github.com/DefinitelyTyped/DefinitelyTyped/commits/HEAD/types/react) Updates `@types/react-dom` from 18.3.7 to 19.2.5 - [Release notes](https://github.com/DefinitelyTyped/DefinitelyTyped/releases) - [Commits](https://github.com/DefinitelyTyped/DefinitelyTyped/commits/HEAD/types/react-dom) Updates `@vitejs/plugin-react` from 6.0.2 to 6.1.0 - [Release notes](https://github.com/vitejs/vite-plugin-react/releases) - [Changelog](https://github.com/vitejs/vite-plugin-react/blob/main/packages/plugin-react/CHANGELOG.md) - [Commits](https://github.com/vitejs/vite-plugin-react/commits/plugin-react@6.1.0/packages/plugin-react) Updates `vite` from 8.0.16 to 8.2.2 - [Release notes](https://github.com/vitejs/vite/releases) - [Changelog](https://github.com/vitejs/vite/blob/main/packages/vite/CHANGELOG.md) - [Commits](https://github.com/vitejs/vite/commits/v8.2.2/packages/vite) --- updated-dependencies: - dependency-name: "@biomejs/biome" dependency-version: 2.5.10 dependency-type: direct:development update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: tsx dependency-version: 4.23.12 dependency-type: direct:development update-type: version-update:semver-minor dependency-group: dev-tooling - dependency-name: typescript dependency-version: 7.0.2 dependency-type: direct:development update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: zod dependency-version: 4.4.3 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@types/node" dependency-version: 26.2.0 dependency-type: direct:development update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@modelcontextprotocol/sdk" dependency-version: 1.30.0 dependency-type: direct:production update-type: version-update:semver-minor dependency-group: dev-tooling - dependency-name: chokidar dependency-version: 5.0.0 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: esbuild dependency-version: 0.28.2 dependency-type: direct:development update-type: version-update:semver-minor dependency-group: dev-tooling - dependency-name: astro dependency-version: 7.2.4 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@astrojs/check" dependency-version: 0.9.10 dependency-type: direct:development update-type: version-update:semver-patch dependency-group: dev-tooling - dependency-name: ws dependency-version: 8.21.3 dependency-type: direct:production update-type: version-update:semver-patch dependency-group: dev-tooling - dependency-name: "@codemirror/lang-markdown" dependency-version: 6.5.2 dependency-type: direct:production update-type: version-update:semver-patch dependency-group: dev-tooling - dependency-name: "@dagrejs/dagre" dependency-version: 3.1.1 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@tiptap/extension-link" dependency-version: 3.30.3 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@tiptap/pm" dependency-version: 3.30.3 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@tiptap/react" dependency-version: 3.30.3 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@tiptap/starter-kit" dependency-version: 3.30.3 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@uiw/react-codemirror" dependency-version: 4.25.11 dependency-type: direct:production update-type: version-update:semver-patch dependency-group: dev-tooling - dependency-name: "@xyflow/react" dependency-version: 12.11.3 dependency-type: direct:production update-type: version-update:semver-patch dependency-group: dev-tooling - dependency-name: markdown-it dependency-version: 15.0.0 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: react dependency-version: 19.2.8 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@types/react" dependency-version: 19.2.18 dependency-type: direct:development update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: react-dom dependency-version: 19.2.8 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@types/react-dom" dependency-version: 19.2.5 dependency-type: direct:development update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: react-router-dom dependency-version: 7.18.2 dependency-type: direct:production update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@types/markdown-it" dependency-version: 14.2.0 dependency-type: direct:development update-type: version-update:semver-minor dependency-group: dev-tooling - dependency-name: "@types/react" dependency-version: 19.2.18 dependency-type: direct:development update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@types/react-dom" dependency-version: 19.2.5 dependency-type: direct:development update-type: version-update:semver-major dependency-group: dev-tooling - dependency-name: "@vitejs/plugin-react" dependency-version: 6.1.0 dependency-type: direct:development update-type: version-update:semver-minor dependency-group: dev-tooling - dependency-name: vite dependency-version: 8.2.2 dependency-type: direct:development update-type: version-update:semver-minor dependency-group: dev-tooling ... Signed-off-by: dependabot[bot] <support@github.com>
dependabot
Bot
added
dependencies
Author
|
This pull request was built based on a group rule. Closing it will not ignore any of these versions in future pull requests. To ignore these dependencies, configure ignore rules in dependabot.yml |
dependabot
Bot
deleted the
dependabot/npm_and_yarn/dev-tooling-cfed3acbf0
branch
October 31, 2026 16:47
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps the dev-tooling group with 28 updates in the / directory:
1.9.42.5.104.22.44.23.125.9.37.0.23.25.764.4.324.13.226.2.01.29.01.30.04.0.35.0.00.24.20.28.26.4.87.2.40.9.90.9.108.21.08.21.36.5.06.5.21.1.83.1.12.27.23.30.32.27.23.30.32.27.23.30.32.27.23.30.34.25.104.25.1112.11.012.11.314.2.015.0.018.3.119.2.818.3.3119.2.1818.3.119.2.818.3.719.2.56.30.47.18.214.1.214.2.06.0.26.1.08.0.168.2.2Updates
@biomejs/biomefrom 1.9.4 to 2.5.10Release notes
Sourced from @biomejs/biome's releases.
... (truncated)
Changelog
Sourced from @biomejs/biome's changelog.
... (truncated)
Commits
05797b1ci: release (#11450)5f1ed34chore: add CodSpeed and Cloudflare to sponsors (#11410)2081460ci: release (#11309)6559e6cfeat: adds theuseAstroClientOnlyDirectiveValuerule for .astro files (#11299)a8798eafeat: useNamedLayer (#11377)85aac73feat(lint): addnoUnsafeTypeAssertionrule (#11221)e65f07efeat(lint): nursery useControlLabel (#11312)ba8aa18feat(lint): adduseTailwindShorthandClasses(#10312)6b8f09cci: release (#11236)23c0369feat(lint): nursery noInvalidPropertyInitValue (#11187)Maintainer changes
This version was pushed to npm by GitHub Actions, a new releaser for
@biomejs/biomesince your current version.Updates
tsxfrom 4.22.4 to 4.23.12Release notes
Sourced from tsx's releases.
... (truncated)
Commits
ed9d330fix: shimimport.metawhen tokens are split by comments or newlines (#829)651f5betest: cover CommonJS TypeScript import.meta pathsbd3bc64test: cover CommonJS loader source fallback55cbecefix: preserve async ESM require fallback6c5ba85docs: document CommonJS default interopec1bcd5fix: support nyc coverage discovery (#710)b6e5b48docs: clarify CommonJS default imports2f55884fix: map Node test locationsde935d5docs: document Node source-map stack formattingb94f46ffix: support data URLs in tsImportUpdates
typescriptfrom 5.9.3 to 7.0.2Release notes
Sourced from typescript's releases.
Commits
1e4744dMerge branch 'main' into ts7-releasea5a219cmicrosoft/typescript-go#4558ecfe30dUpdate status localization5de25b5Hide executable name in TypeScript statusd7ce74aShow bundled TypeScript version for packaged servers29be66aCorrect TS 7 release version to 7.0.2ed2bd1bMerge branch 'main' into ts7-release8873075Bump the github-actions group across 1 directory with 3 updates (microsoft/ty...9427131Set up stable / nightly extension split, other prep (microsoft/typescript-go#...d4eaca5microsoft/typescript-go#4549Maintainer changes
This version was pushed to npm by microsoft1es, a new releaser for typescript since your current version.
Updates
zodfrom 3.25.76 to 4.4.3Release notes
Sourced from zod's releases.
... (truncated)
Commits
1fb56a5docs: document release procedure in AGENTS.mdf3c9ec04.4.3c2be4f8fix(v4): generalize optin/fallback to transform; restore preprocess on absent...1cab693fix(v4): restore catch handling for absent object keys (#5937) (#5939)b8dffe9docs: remove Numeric and Speakeasy (2+ missed monthly cycles)9195250docs: remove Mintlify from bronze sponsors (churned)2c70332docs: normalize bronze sponsor logos to github avatar pattern7391be8docs: prune lapsed silver/bronze sponsors and add active ones2aeec83docs: prune lapsed gold sponsors and rebalance logo sizing4c2fa95docs: use Zernio primary wordmark for gold sponsor logoMaintainer changes
This version was pushed to npm by GitHub Actions, a new releaser for zod since your current version.
Updates
@types/nodefrom 24.13.2 to 26.2.0Commits
Updates
@modelcontextprotocol/sdkfrom 1.29.0 to 1.30.0Release notes
Sourced from @modelcontextprotocol/sdk's releases.
Commits
2d889f2chore: bump version to 1.30.0 (#2563)e3f3daaFix SSE keep-alive timer lifecycle in Streamable HTTP server transport (v1.x)...bb5a718fix(deps): widen@hono/node-serverpast GHSA-frvp-7c67-39w9 (#2549)1dad263fix: send SSE keep-alive comment frames from Streamable HTTP server transport...69749aaValidate Content-Type by parsed media type instead of substring match (v1.x) ...369513dfix: support Zod 3.25 method literals (#2368)e7ee57cv1 stdio buffer limit (#2239)c36e1efAdd end-to-end test suite (#2167)bf1e022chore(ci): switch publish to OIDC trusted publishing (#1839)9edbab7fix(server): prioritize zod issues and format them (#1503)Maintainer changes
This version was pushed to npm by GitHub Actions, a new releaser for
@modelcontextprotocol/sdksince your current version.Updates
chokidarfrom 4.0.3 to 5.0.0Release notes
Sourced from chokidar's releases.
Commits
c0c8d20Release 5.0.0.b211cecRemove src from npm8742246Upgrade dev deps, jsbt, ci files. Upgrade readdirp to v5.de5a34cMerge pull request #1442 from paulmillr/flaky-bunsc08a6c4fix: throttle based on dir + target0c55ab3test: wait for explicit calls in directory testce81be5perf: re-use double slash regex (#1435)7d9c1edMerge pull request #1433 from paulmillr/super-matrices3915541Merge pull request #1430 from paulmillr/esm-only9308bedchore: use Nodejs 24 in CI (#1432)Maintainer changes
This version was pushed to npm by GitHub Actions, a new releaser for chokidar since your current version.
Updates
esbuildfrom 0.24.2 to 0.28.2Release notes
Sourced from esbuild's releases.
... (truncated)
Changelog
Sourced from esbuild's changelog.
Commits
609683dpublish 0.28.2 to npm11b1fe4add to release notesab50d91css: fix green/blue channel swap in oklch gamut mapping (#4488)04627b6fix #4498:asyncTLA checks need a worklist5c15177disablegoplsin thegofolderfc2ee9bcss: adjust parser to allow--foo: {...}209db54release notes for css nesting bugfixc625d31fix #4497: preserve nested ampersands during minification (#4500)34474e2better isolation of current part in js parser07f6e8cfix #4507:importassignment tree-shaking bugMaintainer changes
This version was pushed to npm by GitHub Actions, a new releaser for esbuild since your current version.
Updates
astrofrom 6.4.8 to 7.2.4Release notes
Sourced from astro's releases.
... (truncated)
Changelog
Sourced from astro's changelog.