feat(prompts): add documentation maintenance instructions to agent prompts

[aaight]

Summary

• New documentation-maintenance.eta partial — reusable instructions covering when/what/how to update docs (CLAUDE.md, README, docs/, JSDoc) when code changes affect documented behavior
• 5 code-modifying agent prompts updated — implementation, respond-to-review, respond-to-ci, respond-to-pr-comment, resolve-conflicts all include the new partial after test-protocol
• Review agent gets a new "Documentation" subsection under "What to Verify" (not the partial — reports gaps rather than fixing them)
• Planning agent gets step 6 in Codebase Pattern Analysis to check documentation before writing the plan, with guidance to include doc update steps for changes that affect documented behavior
• Tests — 127 tests pass; new describe blocks cover the partial, all 5 code-modifying agents, the review agent's doc section, and the planning agent's doc guidance

Key decisions

• The partial is action-oriented ("update docs") for code-modifying agents; the review agent gets separate read-only wording ("report gaps, don't fix them") to keep roles clean
• Instructions are framed as "check IF changes affect docs" to avoid unnecessary iterations on internal refactors or test-only changes
• Partial follows the concise style of existing partials (≈35 lines vs test-protocol's 8 lines) with a brief checklist at the end

Test plan

• npm test -- tests/unit/agents/prompts.test.ts — all 127 tests passing
• npm run lint — 0 new errors (2 pre-existing warnings unrelated to this PR)
• npm run typecheck — 0 type errors
• resolve-conflicts added to mock agent types list and all relevant test lists

Trello card: https://trello.com/c/3d75QVrc/589-add-documentation-maintenance-instructions-to-agent-prompts

🤖 Generated with Claude Code

🕵️ claude-code · claude-sonnet-4-6 · run details

[nhopeatall]

Summary

LGTM — Well-structured addition of documentation maintenance instructions to agent prompts with clean role separation.

The implementation is sound:

• Correct agent targeting: The partial is included only in the 5 code-modifying agents (implementation, respond-to-review, respond-to-ci, respond-to-pr-comment, resolve-conflicts). Non-code-modifying agents (splitting, planning, debug, backlog-manager, respond-to-planning-comment) are correctly excluded.
• Role-appropriate wording: The review agent gets read-only "report gaps" instructions separate from the action-oriented partial, maintaining the read-only agent contract. The planning agent gets a single step (step 6) to check docs during pattern analysis, which is the right scope for a planning-only agent.
• Consistent placement: The partial is placed after test-protocol and before verification-protocol in all 5 templates — same ordering pattern.
• Conditional framing: Instructions are framed as "check IF changes affect docs" with explicit skip conditions (internal refactors, test-only changes, bug fixes restoring existing behavior), avoiding unnecessary doc-update churn.
• Tests are thorough: 127 tests covering the partial content, all 5 code-modifying agents, the review agent's doc section, planning agent guidance, and negative assertions (review/planning prompts do NOT contain the partial's checklist).
• CI all green: All 7 checks passing.

Minor note (not blocking): respond-to-pr-comment.eta lacks a trailing newline — this is a pre-existing issue, not introduced by this PR.

🕵️ claude-code · claude-opus-4-6 · run details