feat(agents): expose bounded public work filters

[decode2]

Linked issue

Refs #1702, additive published-work discovery rather than closure of owner decisions.

PR type

• New feature (type:feature)

Summary

• Expose explicit optional orchestrator_list.filter using the tested bounded basic/related work-query library.
• Preserve absent-filter listing and targeted catalog continuation; validate strict arguments before peer/profile discovery.
• Prove classified publication/search, relations, replacements/withdrawals, privacy and scoped effects through production tools and actual SDK contexts.

Changes

Surface	Change
extensions/gentle-agents.ts	Discoverable strict filter schema and explicit work-search envelope
tests/gentle-agents.test.ts	Public RED/GREEN, invalid-before-scan, default and continuation behavior
tests/orchestrator-consultation-sdk.test.ts	Real SDK production publication/filter/related acceptance, local providers and effect counters
assets/orchestrator-delegation.md, docs/gentle-agents-activity.md, odd/tasks/work-discovery.md	Lazy guidance, bounds and evidence

Test plan

• RED: valid filter returned old text rather than expected query JSON; GREEN: one focused test passed.
• Writer public/SDK/query/budget group: 250 passed; broader transport/runner group: 248 passed, overlapping—not summed.
• Independent exact combined suite: 452 passed, zero failed/skipped/cancelled/todo, 14.519s; six source/index hashes unchanged.
• Actual SDK production factories, DefaultResourceLoader and bound contexts; explicit schema-2 publication, AND fields/scoped ref collisions, possible/declared related reasons, ghost source exclusion, replace/withdraw and private-history exclusion.
• Metadata searches add zero owner/nested-helper calls, cost dialogs or caller Git probes; ordinary SDK caller driver takes two local model turns per invocation.
• Existing five consented helper streams unchanged; helper whitelist still text-only.
• Type baseline 186 diagnostics unchanged, not clean compilation; runtime eight unchanged, whitespace and core 8,192-byte guard passed.
• Shellcheck / skill-load testing: N/A, no scripts or skills changed.

Contributor checklist

• Existing approved issue with nonclosing partial-feature reference.
• Exactly one type:feature requested; verify readback.
• Implementation, tests, docs and conventional work-unit commit together.
• No AI attribution or Co-Authored-By trailers.

Chain context

Field	Value
Strategy	Sequential feature-parent stack
Position	5, independently usable public classified-work search
Base	feat/work-related, PR #1786 (c494f726)
Depends on	#1786 → #1785 → #1780 → #1779
Next	Explicit bounded helper classification capture
Review budget	299 additions + deletions / 400, tests/docs/tracker included
Starts at	Complete tested query library
Ends with	Existing public list tool opt-in filter and actual SDK acceptance
Rollback	Remove optional filter registration/guidance/tests; legacy list and query library remain

main
 └── #1779: publication
      └── #1780: allocated-task annotation
           └── #1785: basic query library
                └── #1786: related-source queries
                     └── 📍 this PR: public list / SDK acceptance
                          └── next: helper work capture

No filter means the prior session list, with no bulk curated-note export. Explicit {} indexes classified work. Filters and exact source selection reuse the bounded library: max 64 examined peers, existing incomplete scans fail closed, one catalog page per peer and a 16-KiB query JSON with whole-row omissions and explicit unknown/non-exhaustive coverage. Shared typed refs are descriptive declarations, not executable dependencies or permission; shared tags/categories are possible overlap only.

New task classification is explicit owner data; SDK ghost annotations are historical declarations, never proof of launched children. This fixture proves supported SDK contexts and local deterministic providers, not real OS AgentRunner execution, human consent, TUI or RPC wire-client interaction. Simulated UI caveats and exact owned profile/socket cleanup remain intact. Guidance is lazy-only; no core prompt budget change, cost/admission protocol change, new model/Git probe, store or tool verb.

Out of scope: helper work whitelist, owner-only decisions, native review consumption, main merges and runtime activation. Native assessment remains medium/runtime-large/under-budget with unknown outcome; independent functional proof is not native closure.

Lazy-module CI correction

• Latest head: 63d4b29380cd0a28b8a53eb4c57b451bfde6bb8b; the earlier 452-test independent record above belongs to the original 2e5de12f candidate, not this correction.
• CI exposed the separate lazy delegation cap: 21,595 bytes exceeded 20,000. Guard RED: eight passed/one failed; full-suite packaging RED: 4,780 passed/one failed/44 skipped.
• Full work guidance and JSON examples remain verbatim in the packaged human guide; routing names one canonical writer/explorer rule. Neither the 20,000-byte module cap nor the 8,192-byte core cap was raised.
• tests/task-size-routing-contract.test.ts and tests/package-manifest.test.ts cover preserved guidance, canonical precedence/stop and missing/wrong/dangling references without requiring duplicate prose.
• Writer correction GREEN: six focused suites, 316 passed; full pnpm test, 4,781 passed/zero failed/44 skipped, with provider-contract and runtime-harness checks passing. Runs overlap and are not summed.
• Parent readback verified the reference assertions; type baseline remains 186 recorded diagnostics with no regressions, eight runtime modules match, and package-resource verification passed (159 files, 69 pinned artifacts). Local packed-install verification was not rerun; fresh remote CI is pending.
• Actual delegation asset: 19,496 bytes. Native assessment is medium/runtime-large, under budget, outcome unknown; its plan accepts writer self-verification without a separate verifier. No native closure, main merge, runtime reload or issue closure occurred.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: f70af8a5-a7b2-4dd3-9a06-7258f24d2735

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}