Add Tasks skill for CLI bootstrap and browser plan review

[calvin-archastro]

Review on ArchCode

Problem and author intent

Coding agents can install the public Rooms skill, but have no equivalent Tasks skill explaining how to bootstrap ArchDev and drive a human plan review. The CLI already owns draft validation, revision-aware feedback, approval, and durable task IDs; agents need an executable workflow around those commands.

This adds an independently installable Tasks skill that converts a conversation or harness plan into a dependency graph, opens browser review, revises the same plan from feedback, and verifies that approval saved Tasks.

What changed

• Follow the Rooms distribution pattern with self-contained Unix and PowerShell bootstrap scripts using the pinned official installer. Reuse a capable CLI and update one lacking review commands.
• Teach draft construction, stable node keys, prerequisite edges, unresolved decisions, persistent review processes, browser handoff, feedback cursors, revision fencing, save verification, and receipt-based restart recovery.
• Distinguish transient save failures from external Task edits, which require reconciliation and a newly approved revision.
• Document installation and add a Tasks skill packaging/bootstrap job to the existing installer workflow.

Scope: CLI distribution and agent instructions; no backend or frontend runtime changes. Users can install the skill and ask their coding agent to manage the review loop. The human retains approval of saving Tasks; implementation remains a separate request.

Risk

Low: runtime changes are confined to bootstrap scripts following the existing Rooms pattern. Instructions can still steer incorrect agent behavior, so the example and workflow were checked against the CLI source and independently reviewed.

Testing

• Passed bash tests/tasks-skill.sh. Canonical packaging/bootstrap proof: tests/tasks-skill.sh, the complete script target. It crosses the real skills-manager process boundary for global and repository installs, then invokes the installed bootstrap against a fixture installer/CLI. It asserts an absolute executable path, reuse without installation, upgrade of an outdated CLI, and failure without a misleading successful path. The release/account boundary is substituted; this is not a live Tasks end-to-end proof.
• Automatic coverage: .github/workflows/installer-smoke-test.yml → tasks-skill → bash tests/tasks-skill.sh, on every pull request, main push, and manual dispatch. No Aster target exists in this public distribution repository; the workflow runs the script directly without test filters or skips.
• Passed skill frontmatter validation, Bash syntax checks, and git diff --check.
• Parsed the documented JSON example with the actual shared parseTaskReviewDraft implementation; checked installed CLI 0.35.4 command help and bootstrap reuse.
• Independent source walkthrough covered feedback, revisions, saving, external-edit recovery, and restart identity.

Follow-ups and limitations

The live browser → local review server → authenticated Tasks API journey was not run for this documentation/distribution change. A dedicated test account and browser-connected environment are needed to exercise and automatically cover that boundary. The new CI job verifies installation/bootstrap only. PowerShell bootstrap was reviewed against the Rooms exemplar but not executed locally; it has no dedicated Tasks bootstrap CI coverage yet.