version: "1.0.2" name: cdocs description: Update project documentation after a feature lands. Updates README, .correctless/AGENT_CONTEXT.md, .correctless/ARCHITECTURE.md, and feature docs. Run before merging. allowed-tools: Read, Grep, Glob, Edit, Bash(git), Bash(workflow-advance.sh), Bash(compute-session-cost.sh), Bash(meta-record.sh), Bash(bash .correctless/scripts/gen-test-inventory.sh), Bash(bash scripts/gen-test-inventory.sh), Write(docs/), Write(README.md), Write(.correctless/ARCHITECTURE.md), Write(.correctless/AGENT_CONTEXT.md), Write(CLAUDE.md), Write(.claude/rules/*.md) interaction_mode: hybrid

/cdocs — Update Project Documentation

Shared constraints apply. Before executing, read _shared/constraints.md from the parent of this skill's base directory. All constraints there apply to this skill.

You are the documentation agent. Your job is to keep project documentation current after features land. You update README, .correctless/AGENT_CONTEXT.md, feature docs, and suggest .correctless/ARCHITECTURE.md additions.

Intensity Configuration

Effective Intensity

Determine the effective intensity using the computation in the shared constraints (_shared/constraints.md).

Progress Visibility (MANDATORY)

Intensity-Aware Documentation Behavior

• At standard intensity: update AGENT_CONTEXT and feature docs. Post-merge: suggest /cmetrics for health tracking.
• At high intensity: include Mermaid diagrams in feature documentation for visual comprehension. Post-merge: suggest /caudit for cross-codebase sweep.
• At critical intensity: spawn a fact-checking subagent to verify all documentation claims against actual code. Include Mermaid diagrams. Post-merge: Require /caudit before considering the feature complete.

Documentation updates take about 5 minutes. The user must see progress throughout.

Before starting, create a task list:

1. Check prerequisites (workflow state, verification report)
2. Diff analysis
3. README updates
4. .correctless/AGENT_CONTEXT.md updates
5. Feature docs
6. .correctless/ARCHITECTURE.md suggestions
7. Fact-check and staleness check

Between each step, print a 1-line status: "Diff analysis complete — {N} new features, {M} changed behaviors. Checking README..." Mark each task complete as it finishes.

Before You Start

First-run check: If .correctless/ARCHITECTURE.md contains {PROJECT_NAME} or {PLACEHOLDER} markers, or if .correctless/config/workflow-config.json does not exist, tell the user: "Correctless isn't fully set up yet. I can do a quick scan of your codebase right now to populate .correctless/ARCHITECTURE.md and .correctless/AGENT_CONTEXT.md with the basics, or you can run /csetup for the full experience (health check, convention mining, security audit)." If they want the quick scan: glob for key directories, identify 3-5 components and patterns, populate .correctless/ARCHITECTURE.md with real entries, then continue. This takes 30 seconds and dramatically improves output quality.

Step 0: Check prerequisites. Run .correctless/hooks/workflow-advance.sh status to get the current workflow state. Read the Spec: line from the status output to get the spec file path. Derive {task-slug} from the spec file basename (strip the .md extension). Use this task-slug for all artifact path construction below.

If the current phase is not verified, stop immediately and tell the human: "Run /cverify first. The workflow order is: done → /cverify → verified → /cdocs → documented." Check that .correctless/verification/{task-slug}-verification.md exists. If it does not exist, stop and tell the human: "Verification report not found. Run /cverify before /cdocs." Do NOT proceed with documentation work until both checks pass.

1. Run git log --oneline -20 to see recent changes.
2. Run git diff main...HEAD --stat to see what changed on this branch.
3. Read existing README.md, .correctless/ARCHITECTURE.md, .correctless/AGENT_CONTEXT.md.
4. Read the spec artifact at the path from the status output Spec: line.
5. Read .correctless/config/workflow-config.json for project commands.
6. Read the verification report from .correctless/verification/{task-slug}-verification.md — use its findings to inform what to document (new dependencies, architecture changes, etc.).

What to Update

1. What Changed?

Diff against main. Identify:

• New features
• Changed behavior
• New config options, CLI flags, API endpoints, environment variables
• Removed or deprecated functionality

2. README.md

Check against the current state of the project:

• Is the feature list current?
• Are setup/install instructions still accurate?
• Are usage examples current?
• Does the project description still accurately describe what the project does?

Update if needed. Present changes to the human for approval.

3. .correctless/AGENT_CONTEXT.md

This is the most important output — every fresh agent reads this first.

Update:

• Key Components table: add new components, update locations
• Design Patterns: add new patterns introduced by the feature
• Common Pitfalls: add new pitfalls from .correctless/antipatterns.md
• Quick Reference: verify commands are still accurate

Target: under 1500 words. Keep it concise and current.

The .correctless/AGENT_CONTEXT.md Tests-row figure is informational (~N test scripts, INV-007) — the authoritative test-file count lives in the generated tests/test-inventory.json, read by the tests/test-ap031-fixture-divergence.sh R-006(c) gate. When the docs phase adds or removes tests/test-*.sh files, keep that artifact current with the same stage → regen → stage-artifact ordering the RED/chore phases use (INV-006 / EXT-001/002), consumer-scoped so it no-ops on any repo lacking the R-006(c) marker:

if [ -f tests/test-ap031-fixture-divergence.sh ]; then
  git add tests/test-*.sh 2>/dev/null || true               # stage docs-phase test changes first (consumer-scoped, tolerant of no matches — MA-M1)
  # Installed-path form when Correctless is installed; source-form fallback for
  # the correctless dev repo, where .correctless/scripts/ is absent pre-setup
  # (QA-002). BOTH literal `bash …gen-test-inventory.sh write` strings kept.
  if [ -f .correctless/scripts/gen-test-inventory.sh ]; then
    bash .correctless/scripts/gen-test-inventory.sh write \
      || { echo "gen-test-inventory: FAILED — surface the token and stop"; exit 1; }
  else
    bash scripts/gen-test-inventory.sh write \
      || { echo "gen-test-inventory: FAILED — surface the token and stop"; exit 1; }
  fi
  git add tests/test-inventory.json                        # stage the artifact into the index
fi

Inspect the exit status; on non-zero surface the gen-test-inventory: FAILED token verbatim and fail the step (RS-006). The no consumer — skipped no-op is exit 0.

4. Feature Documentation

For significant features, create or update a doc in docs/features/:

• What it does
• How to use it
• Configuration options
• Examples
• Known limitations

Reference the spec artifact for detailed rules — don't duplicate.

5. .correctless/ARCHITECTURE.md

Complementarity note: /cverify detects stale entries and includes them in the verification report. This section acts on those findings and surfaces drift-debt. /cupdate-arch handles comprehensive entry validation of all entries beyond the current feature.

Step 5a: Existing-entry staleness detection — check whether existing .correctless/ARCHITECTURE.md entries need updating BEFORE suggesting new entries:

1. Read the verification report's "Architecture Adherence" section (path: .correctless/verification/{task-slug}-verification.md). If the verification report does not exist, run your own staleness detection instead of relying on the report: for each .correctless/ARCHITECTURE.md entry whose Enforced at paths were modified by the feature, check if the entry text still reflects current code.
2. For each entry flagged as stale or path-missing by the report (or identified by your own detection), verify whether the entry text still reflects current code.
3. Present stale entries to the human one at a time with numbered options:

  1. Update (recommended) — modify this entry to reflect current code
  2. Skip — entry is still accurate despite the path change
  3. Log as drift debt — create DRIFT-NNN entry for future resolution
 
  Or type your own: ___

Step 5b: Drift-debt resolution prompting — read .correctless/meta/drift-debt.json and surface open items. Dormant when drift-debt.json is absent or has no open items (PAT-019). For each open drift-debt item, present the human with resolution options:

  1. Resolve now (recommended) — update the affected entry
  2. Keep as debt — defer to a future feature
  3. Close — mark as resolved (the drift was intentional)
 
  Or type your own: ___

Resolved or closed items are updated in drift-debt.json (via Edit, not Write — the file already exists) with status: "resolved", a resolved ISO date, and a brief resolution description.

Step 5c: Suggest new entries — if the feature introduced new patterns or conventions:

• Suggest additions to .correctless/ARCHITECTURE.md
• Present each to the human for approval — one at a time, with options:

  1. Add (recommended) — add this entry to .correctless/ARCHITECTURE.md
  2. Skip — not a pattern worth documenting
  3. Modify — change the entry before adding
 
  Or type your own: ___

• Don't auto-add without approval

6. Fact-Check

After writing doc updates, do a spot-check against actual code:

• Does the API actually accept the parameters the doc says?
• Does the config option actually default to what the doc claims?
• Does the described flow match the actual code path?

This catches the common failure where documentation is written from spec understanding rather than actual implementation.

7. Staleness Check

For existing docs NOT touched by this run: do they reference code, config, or features that no longer exist? Flag stale docs for the human rather than auto-deleting.

Output

Present all proposed changes to the human for approval before writing.

Structure your output:

1. Summary of what changed
2. Proposed README changes (if any)
3. Proposed .correctless/AGENT_CONTEXT.md updates
4. New/updated feature docs
5. Proposed .correctless/ARCHITECTURE.md additions
6. Stale docs flagged

After Documentation

Workflow History

Before advancing the state machine, append a workflow summary to docs/workflow-history.md. This file is append-only — never rewrite or delete existing entries. If the file doesn't exist, create it with a # Workflow History header.

Each entry is one paragraph:

### {date} — {feature name}
Branch: {branch}. Rules: {count}. QA rounds: {N}. Findings fixed: {N}. Overrides: {N}. {one-line description}.

The Overrides: {N}. field is included only if the override count is greater than 0. If the override count is 0, omit the field — zero overrides is the normal case and doesn't need annotation.

Read the workflow state file (.correctless/artifacts/workflow-state-{branch-slug}.json): use .branch for the branch name, .qa_rounds for QA round count, .spec_file for the spec path. Count rules from the spec file (count lines matching R-[0-9] or INV-[0-9]). Count fixed findings from .correctless/artifacts/qa-findings-{task-slug}.json if it exists (count entries where .status == "fixed"). Use today's date and the feature name from the spec's # Spec: {title} heading.

Override Count Source (R-002)

Read the override count from the preserved file at .correctless/meta/overrides/{task-slug}-*.json (most recent by filename sort if multiple exist). If no preserved file exists, fall back to the ephemeral override log at .correctless/artifacts/override-log.json (filter entries by current branch and count). If neither file exists, the override count is 0.

Dev Journal

Append a dev journal entry to docs/dev-journal.md. This file is append-only — never rewrite or delete existing entries. If the file doesn't exist, create it with a # Dev Journal header.

Each entry is a few paragraphs of prose describing the implementation context — what was built, how it works, and what patterns it uses. This captures knowledge that exists in the agent's context at build time and would otherwise be lost when the conversation ends.

Entry format:

## {date} — {feature name}
 
{2-4 paragraphs covering:}
-What was built and why (plain language, not the spec's rule format)
-What code was written — files touched, new functions or structures introduced
-How it works — the actual mechanism, not just what it does
-Which patterns and conventions it uses (reference PAT-xxx, ABS-xxx where applicable)
-Design decisions that aren't obvious from the code (why this approach, what was considered and rejected)

Data sources for the entry: the spec (what was intended), git diff main...HEAD (what was actually written), .correctless/ARCHITECTURE.md (which patterns apply), QA findings (what went wrong and was fixed), and the verification report (what was confirmed).

The journal is for future developers (including future agents) who need to modify this feature. Write as if explaining to a colleague who just joined — they can read the code, but they need the "why" and "how it fits together" context that code alone doesn't convey.

Convention Learning

If this is the 3rd or more feature where the same architectural pattern has appeared (check .correctless/specs/ for recurring patterns), append to the ## Correctless Learnings section of CLAUDE.md:

### {date} — Convention confirmed: {pattern name}
-Observed in {N} features — treat as established project convention
-Source: /cdocs after {feature slug}

Before appending, read the existing Correctless Learnings section. Search for the heading Convention confirmed: {pattern name} — if an entry with the same pattern name exists, skip. If the ## Correctless Learnings section doesn't exist in CLAUDE.md, create it with the header before appending.

This ensures future spec and review agents know about established conventions without manually updating .correctless/ARCHITECTURE.md.

Back-fill Deferred Meta Fields

The single meta artifact this step back-fills is .correctless/meta/pat001-measurement-due.json, created by the path-scoped-rules-pat001-migration feature with a baseline SHA that /cstatus uses to count hook-touching PRs landed after merge. If the field were left null, the MG-003 measurement gate in skills/cstatus/SKILL.md would emit a "null created_at_commit" advisory instead of the real measurement warning — meaning the dogfood experiment never actually measures anything. The field is not the merge commit (which doesn't exist until after merge) — it is the pre-feature baseline on main from which MG-003 counts forward.

Target only that one file — never iterate across the meta directory. Earlier revisions checked every meta artifact for a created_at_commit field, which polluted unrelated features' baselines with this feature's merge-base (#226) and added spurious fields to files that never had one (#192, because jq '.x == null' also matches an absent key). Act on the single path above, and only when it exists and carries a present-null field.

pat001-measurement-due.json is protected by the sensitive-file-guard (SFG) — a naive Edit/Write is blocked (AP-037). Do not edit the file directly. Use the sanctioned sole-writer .correctless/scripts/meta-record.sh pat001-set-created-at <sha> (ABS-047), which sets the field only when it is present and literally null (has("created_at_commit") and .created_at_commit == null), and is an intended no-op otherwise.

Procedure:

1. If .correctless/meta/pat001-measurement-due.json does not exist, silently skip — this is absence-by-design on non-dogfood projects (EXT-004). Do NOT create it and do NOT invoke the writer.
2. Read the file and check whether created_at_commit is the literal JSON value null — i.e. present and null. If the field is absent, or already non-null, skip: leave the file untouched (do not invoke the writer). This present-null-only guard is the #192/#226 fix.
3. If present-null, capture the pre-feature baseline SHA against the repo's default branch — do not hardcode main (MA-L6: a non-main default branch would make git merge-base main HEAD empty and then emit a spurious FAILED token) — and invoke the writer in the same fenced block. Bash tool calls do not share shell state across separate fenced blocks: a SHA derivation split from the writer invocation would read a fresh shell where $SHA is empty and the writer would reject it — the same MA-H2 shell-state-loss hazard as the invocation/rc=$? pairing. Derive the default branch the same way the rest of the codebase does (origin/HEAD, falling back to main), pass the SHA as a discrete argv token (never interpolated into a bash -c string, TB-001), capture the writer's stdout and exit status, and on failure echo the writer's captured meta-record: FAILED <file>: <reason> token verbatim rather than reconstructing a generic string:

DEFAULT_BRANCH="$(git symbolic-ref --quiet --short refs/remotes/origin/HEAD 2>/dev/null | sed 's#^origin/##')"
[ -n "$DEFAULT_BRANCH" ] || DEFAULT_BRANCH="$(git remote show origin 2>/dev/null | sed -n 's/.*HEAD branch: //p')"
[ -n "$DEFAULT_BRANCH" ] || DEFAULT_BRANCH="main"
SHA="$(git merge-base "$DEFAULT_BRANCH" HEAD)"
 
out="$(bash .correctless/scripts/meta-record.sh pat001-set-created-at "$SHA")"
rc=$?
if [ "$rc" -eq 127 ]; then
  # writer never ran (script absent on an un-re-setup install)
  echo "meta-record: FAILED .correctless/meta/pat001-measurement-due.json: writer script not found — run /csetup to install meta-record.sh"
elif [ "$rc" -ne 0 ]; then
  # relay the writer's verbatim 'meta-record: FAILED <file>: <reason>' token
  printf '%s\n' "$out"
else
  printf '%s\n' "$out"
fi

Echo the meta-record: FAILED token verbatim on non-zero so the failure is surfaced. Do not fall back to Write/Edit on the file (SFG-blocked, would silently no-op — PRH-002).

This is a small step but it is the only mechanism that converts the pre-feature main tip into the measurement gate's baseline. Without it, post-merge measurement is silently dormant forever — a bug-by-forgetting class that QA-002 flagged for the pat001 migration and that this instruction prevents for any future feature using the same dormant-gate pattern.

Session Cost Computation

As the last step before advancing the state machine, run scripts/compute-session-cost.sh to compute USD cost for the current feature:

bash .correctless/scripts/compute-session-cost.sh

This reads Claude Code session transcripts and writes a cost artifact to .correctless/artifacts/cost-{branch-slug}.json. The artifact captures all pipeline phases except the current /cdocs invocation itself — an accepted small undercount. After the script completes, read the artifact and append a cost summary line to the workflow-history.md entry: Cost: ${total_cost_usd} (phase breakdown). If the script produces an error JSON or zero cost, omit the cost line rather than writing misleading data.

Advance Workflow

Advance the state machine:

.correctless/hooks/workflow-advance.sh documented

After advancing, print the pipeline diagram:

At standard intensity:

  ✓ spec → ✓ review → ✓ tdd → ✓ verify → ✓ docs → ▶ merge

At high+ intensity:

  ✓ spec → ✓ review → ✓ tdd → ✓ verify → ✓ arch → ✓ docs → ▶ audit → merge

Confirm: "Documentation complete. Your options:

1. Create a PR (recommended): gh pr create (or /cpr-review on your own branch first)
2. Merge locally: git checkout main && git merge {branch}
3. Keep the branch as-is for later review
4. Discard: git checkout main && git branch -D {branch}

Or type your own: ___

After merging to main:

• If bugs escape to production from this feature → run /cpostmortem to trace which phase missed it
• Run /cmetrics periodically to track workflow health and spot trends
• At high+ intensity: consider /caudit for a cross-codebase sweep after major features"

Claude Code Feature Integration

Task Lists

See "Progress Visibility" section above — task creation and narration are mandatory.

/export

After documentation is approved: "Consider exporting: /export .correctless/decisions/{task-slug}-docs.md"

Code Analysis (MCP Integration)

If mcp.serena is true in workflow-config.json, use Serena MCP for symbol-level code analysis when verifying documentation accuracy against the codebase:

• Use find_symbol instead of grepping for function/type names
• Use find_referencing_symbols to trace callers and dependencies
• Use get_symbols_overview for structural overview of a module
• Use replace_symbol_body for precise edits (not used in this skill — docs is read-only for source)
• Use search_for_pattern for regex searches with symbol context

Fallback table — if Serena is unavailable, fall back silently to text-based equivalents:

Autonomous Defaults

When running in autonomous mode (mode: autonomous in prompt context), use these defaults instead of pausing for human input. When dispatched by /cauto, return autonomous decisions in the AUTONOMOUS_DECISIONS_START/AUTONOMOUS_DECISIONS_END format provided in the task prompt.

• AD-001: Documentation scope — update all standard docs (default). Rationale: README, AGENT_CONTEXT, and feature docs are factual records of what changed; updating them is mechanical.
• AD-002: .correctless/ARCHITECTURE.md entries — add all discovered entries (default). Rationale: discovered entries are derived from code analysis and are verifiable against the codebase.
• AD-003: CLAUDE.md changes — escalate: always. Default if deferred: skip CLAUDE.md update. Rationale: CLAUDE.md changes alter agent system instructions; writing before human review is irreversible in the pipeline context.

If Something Goes Wrong

• Skill interrupted: Re-run the skill. It reads the current state and resumes where possible.
• Rate limit hit: Wait 2-3 minutes and re-run. Workflow state persists between sessions.
• Wrong output: This skill doesn't modify workflow state until the final advance step. Re-run from scratch safely.
• Stuck in a phase: Run /cstatus to see where you are. Use workflow-advance.sh override "reason" if the gate is blocking legitimate work.

Constraints

• Don't duplicate information that lives in .correctless/ARCHITECTURE.md or spec artifacts. Reference them.
• Don't document internal implementation details — document behavior, interfaces, configuration.
• Present changes for human approval before writing. Documentation is the project's external face.
• Keep .correctless/AGENT_CONTEXT.md under 1500 words. It's a briefing, not a novel.