AIP-19: Build review-session-context skill for .dev-team-review.md

[jodavis-claude]

Work item: AIP-19

AIP-19 required building review-session-context: a new skill that computes, initializes, reads, and writes .dev-team-review.md — a small, worktree/clone-local state file for an in-progress PR review session — using the same sentinel-section append/replace conventions as use-context-file, but independent of it.

Changes

• plugins/dev-team/skills/review-session-context/SKILL.md (new) — documents resolving, initializing, reading, and writing .dev-team-review.md: path resolution at <worktree-or-clone-root>/.dev-team-review.md (no repo-slug, no ~/.dev-team/DEV_TEAM_STATE_DIR), the frontmatter field table, the same sentinel-section append/replace format as use-context-file, the three reserved section names (Specialist Report, Quiz, Publish Log) documented but left unwritten, and a note that this skill is the file's sole writer.
• plugins/dev-team/skills/review-session-context/assets/review_context_template.md (new) — frontmatter template with the ten documented fields (repo_scope, isolation_kind, owner, repo, pr_number, pr_url, worktree_path, head_ref, guide_path, work_item_id), all blank, plus an HTML-comment block listing the three reserved section names for later deliverables.
• plugins/dev-team/skills/review-session-context/scripts/compute-review-context.py (new) — given a worktree/clone root, prints the absolute path to .dev-team-review.md under it. No git-remote/repo-slug logic, unlike use-context-file's compute-context-file.py.
• plugins/dev-team/skills/review-session-context/scripts/init-review-context.py (new) — idempotent creation of the review context file from the template if it doesn't already exist; creates parent directories as needed; exits non-zero (leaving no partial file) on error.
• plugins/dev-team/.claude-plugin/plugin.json (modified) — bumped version from 1.4.1 to 1.4.2, required for any PR touching plugins/<name>/ per this repo's CI convention.

Design decisions

• No dedicated pytest suite for init-review-context.py/compute-review-context.py. The spec classifies review-session-context as Wrapper tier (visual inspection sufficient, no dedicated unit test required by default) and its own isolation-pattern note lists only resolve_pr_reference.py, decide_cleanup_action.py, gather-pr-evidence, and write-review-guide as needing a verification harness. Verified the scripts manually instead: fresh creation, idempotency (second run is a byte-for-byte no-op), the error path (unwritable parent exits non-zero via caught OSError, no partial file/directory left behind), and the sentinel append/replace format itself.
• No E2E/Gherkin scenarios written — zero .feature files exist anywhere in this repo, and this component introduces no new externally-observable entry point of its own; the actual /dev-team:review-pr command that will call it is a separate, later task (AIP-23).
• init-review-context.py takes only the target <review-context-file> path as its argument, since (unlike use-context-file's template) none of .dev-team-review.md's ten frontmatter fields need substitution at init time — every field starts blank and is filled in later by whichever component resolves it.
• Plugin version bump included in the same commit as the new skill files, since it's small, mechanical, CI-required glue work with no test implications of its own.

Testing completed

Manual verification of init-review-context.py and compute-review-context.py: fresh creation (frontmatter fields present and blank, correct order), idempotency (second run is a byte-for-byte no-op), the error path (unwritable parent path exits non-zero, leaves no partial file/directory), and the sentinel section append/replace format (appended a new section, then replaced its content, both succeeding as documented).

[jodavis-claude]

Reviewed AIP-19: Build review-session-context skill for .dev-team-review.md.

All three exit criteria are met:

• Computes and initializes .dev-team-review.md with the documented frontmatter fields (repo_scope, isolation_kind, owner, repo, pr_number, pr_url, worktree_path, head_ref, guide_path, work_item_id), matching the spec's Interfaces section exactly.
• Writing a named section uses the identical sentinel-section append/replace format documented by use-context-file.
• The three reserved section names for later deliverables (Specialist Report, Quiz, Publish Log) are documented in the initial template and left unwritten.

Verified manually: fresh creation, idempotent second run, and the error path (unwritable parent exits non-zero via a caught OSError, no partial file left behind) all behave as documented. plugins/dev-team/.claude-plugin/plugin.json version was correctly bumped (1.4.1 -> 1.4.2) per this repo's CI convention for any PR touching plugins/<name>/.

No Priority 1-4 issues found (correctness/fault-tolerance, security, performance, documentation).

Minor (non-blocking) note: init-review-context.py/compute-review-context.py have no dedicated pytest suite, unlike the mirrored reference init-context-file.py/test_init_context_file.py. This follows the spec's explicit Wrapper-tier classification for this component (visual/manual inspection sufficient) rather than the Testable-tier pattern the reference happens to use, so I'm not blocking on it, but flagging since a future reviewer may want to weigh in if this component's scope grows.

Approving.

[jodavis-claude]

Sign-off review for AIP-19 (review-session-context skill).

Prior review threads: none existed — the first-pass review approved with no inline comments and zero open threads. Nothing to resolve here.

New commits since first-pass review: none. The branch is unchanged since that review (single commit c43be32), so there is nothing new to scan.

Re-verified: SKILL.md, compute-review-context.py, init-review-context.py, and review_context_template.md against the task brief's exit criteria, _spec_ReviewPrGuide.md's Interfaces section, CONTRIBUTING.md, and the mirrored reference (use-context-file). All three exit criteria remain met, the sentinel-section format matches, reserved section names (Specialist Report, Quiz, Publish Log) are documented and left unwritten, error handling in init-review-context.py is caught and non-silent, and plugin.json's version bump (1.4.1 -> 1.4.2) is present. No Priority 1-4 issues found.

Signing off.