Wrapup

Called automatically by sprint complete — do not invoke directly.

Runs after a session, feature, bug fix, or ticket closes. Skip steps that do not apply.

Every Read skills/... reference below is a plain relative path inside canon's own repo. In a consumer project, resolve via .claude/skills/... (Claude Code) or .agents/skills/... (Codex/Pi) instead — canon's install symlinks into its real skills/ tree. Known gap: if .claude/skills already existed as a real directory before skills.sh add ran, this file itself (a hidden sub-skill) may not be reachable via .claude/skills — .agents/skills is unaffected (t-1b2c).

Pipeline

code-simplifier → code-reviewer → security-review → doc-audit → refresh docs

Commit & Push belongs to sprint complete's own step 10, not this pipeline — see skills/sprint/reference/complete.md.

code-simplifier/code-reviewer run inline, in-session — scope is "code touched this session" from working memory, no git command needed. security-review also runs inline but derives scope from git: an explicit Base ref (headless CI dispatch grading an existing PR/diff) uses git diff --name-only <base-ref> HEAD; otherwise (normal interactive close) git diff --name-only $(git merge-base HEAD origin/main) HEAD — same branching as shared-gate-protocol.md ## Base-ref derivation and security-review.md's own Scope section. reviewer/evaluator (fresh-subagent gates in skills/sprint/reference/complete.md, not this pipeline) always derive from git — they have no session memory at all.

Skip Logic

Trivial change (single-line, doc-only, mechanical rename): skip all steps except Refresh docs (Commit & Push isn't part of this pipeline — see above). This global clause overrides every per-gate skip criterion below — e.g. a one-line rename in a user-facing doc skips doc-audit via this clause.

Tier: bugfix and demo: true also override the per-gate criteria below wholesale, the same way — they are defined in skills/sprint/reference/complete.md, not here, and this file has no independent knowledge of them: bugfix (complete.md's "Bugfix tier — lighter wrapup") keeps security-review/inline code-reviewer, may skip code-simplifier/doc-audit; demo (complete.md's "Demo mode") keeps only security-review, skips every other gate in this pipeline. Reading this file in isolation without checking plan.md's Tier: or ticket.md's demo: first will apply the wrong gate set on either path.

Before each step, assess and skip if criteria apply — state why in one line, naming which clause justified it (global override or the gate's own criteria) when the two could seem to conflict.

Skip code-simplifier if:

• Change is a single line or a trivial rename
• Change is docs, comments, or config only

Skip code-reviewer if:

• Change is a single-line fix with no design implications
• Change is purely mechanical (rename, format, move file)

Skip security-review if:

• No security-sensitive files or patterns changed
• Security-sensitive means: authentication, authorization, DB queries, externally-facing/untrusted input handling (web forms, API payloads, network requests — not trusted local CLI args a single developer runs against their own machine), file I/O, API endpoints, crypto, session management, environment/secret access, MCP server config or agent tool/permission definitions, CI workflow / GitHub Actions config (.github/workflows/*, or a workflow template the project generates), hardcoded credential/secret access, destructive/irreversible client-side actions (delete-all, bulk clear, cache flush), third-party data egress (adding an analytics/tracking pixel, a third-party script/tag/SDK, or any code that sends PII/credentials/tokens to a third-party endpoint — including from inside an authenticated flow)

Skip doc-audit if:

• No user-facing docs changed and no skill/standards frontmatter changed (so the model pass and tools/doc-check both skip) — matches doc-audit/SKILL.md's own scope: README, docs/*.md, tools/*.md, or description/summary frontmatter in skills/*/SKILL.md/standards/*.md

Steps

1. Read skills/wrapup/gates/simplifier.md, then apply the simplifier to code touched in this session.
2. Read skills/wrapup/gates/reviewer.md, then apply the code-reviewer across all eight dimensions; defer deep security analysis to Step 3.
3. Read skills/wrapup/gates/security-review.md, then apply the security review. Optional scanners can inform the review, but their absence does not skip the gate.
4. Read skills/doc-audit/SKILL.md, then apply the doc audit (tools/doc-check first, then the judgment pass). Do not write to doc-findings.md without explicit confirmation. Fix command accuracy issues before committing. A file this step edits must be listed in plan.md ## Files in the same step — the reviewer grades scope against that list, and a close-time doc edit missing from it was flagged in four sprints (t-b999, t-533f, t-824e, t-bdce).
5. Read tools/handoff.md — plain relative path inside the canon repo itself, or via command -v sprint's containing directory inside a consumer project where tools/ isn't symlinked in (or where sprint on Windows if command -v returns nothing) — then refresh docs:

Review every documentation file touched or referenced during this session and patch anything stale.

Scope (check each that exists):

• DECISIONS.md — any new decisions made this session not yet logged?
• HANDOFF.md — refresh the narrative per the handoff protocol: Current Focus, In Progress, and Next Steps. ## Discoveries is owned by capture — leave it. Decisions belong in DECISIONS.md, not here. Then check size within the <!-- canon:handoff:BEGIN -->/<!-- canon:handoff:END --> markers only — anything outside them may be the user's own content and must never be read, counted, or edited by this step. First confirm both markers actually exist (grep -c '<!-- canon:handoff:BEGIN -->' HANDOFF.md and the same for :END, each must be ≥1) — the range awk silently returns 0 lines (a false "under budget") when a marker is missing, so a count without this guard is meaningless; if either marker is absent, restore the pair around the canon-managed block before counting. Then: awk '/<!-- canon:handoff:BEGIN -->/,/<!-- canon:handoff:END -->/' HANDOFF.md | wc -l. If over 80, prune per tools/handoff.md's "When to Prune" section before finishing this step.
• AGENTS.md / CLAUDE.md — any convention-level learnings to surface? (propose + confirm before writing)
• README — does it document any changed APIs, behaviors, or install steps?
• plan.md (sprint tickets only) — is the whole doc still within the ~500-word budget noted at the top of the file? If it grew past budget during implementation, compress the prose (short bullets, file:line references, no restated history) before closing.
• Any other .md files explicitly opened or modified during the session

Patch stale lines only. Skip files where nothing changed.

Final Output

Report only what matters:

• Changed files and user-visible impact
• Tests run and result
• Critical findings, if any
• Follow-up captured in HANDOFF.md, if any

Address criticals before committing. Improvements are discretionary.

Committing and closing the ticket are the caller's job, not this pipeline's — see the Pipeline note above.