feat(agent): add portable learning capture gate

[ss-o]

Summary

• require a lightweight learning and reuse review before non-trivial work is
  claimed complete
• add one canonical learning-capture runbook and an optional advisory skill
• route and validate both surfaces through the public instruction manifest
• keep No durable learning as a valid silent outcome
• preserve explicit authority for memory, issue, policy, and documentation
  writes

This implements the policy portion of #482. The issue remains open for its
30-task or four-week measurement pilot.

Instruction impact review

1. This is shared mandatory policy plus canonical runbook detail and one
   optional advisory skill.
2. Codex, Claude Code, Copilot, and humans receive the baseline and runbook.
   Gemini CLI is deprecated and is not a verification target.
3. AGENTS.md owns the mandatory gate,
   runbooks/learning-capture.md owns workflow detail, and the skill remains
   advisory.
4. No competing learning log or memory owner is introduced.
5. The public manifest inventories the runbook and skill. Private routing is a
   separately owned delivery step.
6. Mandatory behavior reaches supported runtimes without relying on the skill
   or a hook.
7. Public manifest tests, policy validation, and the instruction size limit
   pass.

Validation

• python3 -B scripts/validate-agent-policy.py
• python3 -B -m unittest scripts/test_validate_agent_policy.py -v
  (70 tests passed)
• trunk check --no-fix --force on all six changed files
• git diff --check origin/main...HEAD

Runtime discovery evidence

• Codex: verified in a fresh task.
• GitHub Copilot CLI: verified through direct repository skill discovery with
  tools disabled.
• Claude Code: unverified because the bounded check exceeded its budget.
• Gemini CLI: deprecated and not invoked.

Safety boundaries

• no automatic memory, issue, ADR, runbook, instruction, skill, or
  documentation write path
• no Codex hook
• no private workspace paths or data

[Copilot]

Pull request overview

Adds an explicit “learning capture” completion gate to the org’s agent policy, with a canonical runbook and an optional Copilot-only advisory skill, all routed via the public instruction manifest and covered by policy validation tests.

Changes:

• Introduce runbooks/learning-capture.md as the canonical workflow for learning/reuse review before declaring non-trivial work complete.
• Update AGENTS.md to require the learning capture review while keeping “No durable learning” as a valid silent outcome.
• Register the new runbook and advisory skill in .github/instruction-surfaces.json and add a unit test to assert their manifest declarations.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
scripts/test_validate_agent_policy.py	Adds assertions that the new learning-capture runbook surface and the advisory skill are declared in the public manifest.
runbooks/learning-capture.md	New canonical runbook defining the learning-capture workflow, routing/destination rules, and authority boundaries.
AGENTS.md	Makes learning capture a mandatory completion step for non-trivial work, delegating detailed workflow guidance to the runbook.
.github/skills/review-project-learning/SKILL.md	Adds an optional Copilot skill that guides users through the learning-capture workflow without expanding authority.
.github/README.md	Documents the new runbook and skill as part of the instruction architecture.
.github/instruction-surfaces.json	Registers the new runbook surface (required) and the advisory skill surface (optional) for manifest-based routing.