| name | doc-all-projects |
| description | Sweep all registered projects and regenerate their user-facing feature documentation if stale. Reads ~/.claude/my-projects.yaml, dispatches one parallel agent per project to run the project-documenter skill, commits results, writes per-project reports to Notes/, and displays a sorted summary. Use when you want to refresh docs across all your projects at once. |
/doc-all-projects
Sweep all registered projects and regenerate their user-facing feature documentation if
stale. Uses parallel agents — one per project. Read the full instructions below and follow
them exactly.
Step 1: Read the project list
Read ~/.claude/my-projects.yaml using Bash:
cat ~/.claude/my-projects.yaml
Parse it with Python to get the list of projects:
python3 -c "
import yaml, json, os
with open(os.path.expanduser('~/.claude/my-projects.yaml')) as f:
data = yaml.safe_load(f)
print(json.dumps(data['projects'], indent=2))
"
You now have a list of {name, path} pairs.
Expected my-projects.yaml format:
projects:
- name: ProjectOne
path: /absolute/path/to/project-one
- name: ProjectTwo
path: /absolute/path/to/project-two
Step 2: Dispatch parallel agents
Using the Task tool, dispatch one general-purpose agent per project in a single
message (all in parallel). Use mode: bypassPermissions.
Fill in {NAME} and {PATH} for each project from the list you parsed above.
Agent prompt template
Copy this template for each project, substituting {NAME} and {PATH}:
You are running the documentation sweep for the {NAME} project at {PATH}.
Use absolute paths for all operations. Never use cd.
Step A — Verify the project
Run:
test -d {PATH}/.git && echo "IS_GIT_REPO=yes" || echo "IS_GIT_REPO=no"
If IS_GIT_REPO=no or the path does not exist, print exactly:
RESULT:
NAME={NAME}
STATUS=Error
FILES=0
COMMIT=—
REPORT=—
ERROR=path not found or not a git repo
Then stop.
Step B — Staleness check
Run each check in order. Stop at the first that sets STALE=true.
Check 1 — Docs/Features/ directory missing:
test -d {PATH}/Docs/Features && echo "DIR_EXISTS=yes" || echo "DIR_EXISTS=no"
If DIR_EXISTS=no → STALE=true.
Check 2 — No .md files in Docs/Features/:
ls {PATH}/Docs/Features/*.md 2>/dev/null | wc -l
If count is 0 → STALE=true.
Check 3 — Docs older than 30 days:
DOC_TS=$(git -C {PATH} log -1 --format="%ct" -- Docs/Features/ 2>/dev/null)
CUTOFF=$(date -d "30 days ago" +%s 2>/dev/null || date -v-30d +%s 2>/dev/null)
echo "DOC_TS=$DOC_TS CUTOFF=$CUTOFF"
If DOC_TS is empty or DOC_TS < CUTOFF → STALE=true.
Check 4 — Source files newer than docs:
SRC_TS=$(git -C {PATH} log -1 --format="%ct" -- . ':(exclude)Docs/' ':(exclude)Notes/' 2>/dev/null)
DOC_TS=$(git -C {PATH} log -1 --format="%ct" -- Docs/Features/ 2>/dev/null)
echo "SRC_TS=$SRC_TS DOC_TS=$DOC_TS"
If SRC_TS > DOC_TS → STALE=true.
If no check matched → STALE=false.
Step C — Skip if not stale
If STALE=false, skip directly to Step F (write a "Skipped" report) then Step H.
Step D — Run project-documenter
Read the project-documenter skill instructions:
cat ~/.claude/skills/project-documenter/SKILL.md
Follow its full Phase 1–4 process using {PATH} as the project root.
Use absolute paths for all file reads and writes.
Track every file written under {PATH}/Docs/Features/ for the report.
Step E — Commit the docs
git -C {PATH} add Docs/
git -C {PATH} diff --cached --stat
git -C {PATH} commit -m "docs: regenerate feature documentation [doc-all-projects]
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>"
COMMIT_HASH=$(git -C {PATH} rev-parse --short HEAD)
echo "COMMIT=$COMMIT_HASH"
If nothing was staged (no changes), set COMMIT_HASH=— and note "No changes to commit".
Step F — Write the report
REPORT_DATE=$(date '+%Y-%m-%d')
REPORT_TIME=$(date '+%Y-%m-%d %H:%M')
mkdir -p {PATH}/Notes
Write a markdown file to {PATH}/Notes/doc-report-$REPORT_DATE.md with this structure:
# Documentation Report — {NAME}
Generated: $REPORT_TIME
## Status
- **Action**: Ran ← or "Skipped"
- **Reason**: Source files newer than docs ← or actual reason
## Files Generated/Updated
- Docs/Features/example.md (new, 4.2 KB)
← list each file actually created or modified, with size
## Commit
`$COMMIT_HASH` — docs: regenerate feature documentation [doc-all-projects]
← use — if skipped or nothing staged
## Notes
← any warnings, errors, or special cases; "none" if clean
Step G — Update .gitignore
grep -q 'doc-report-\*\.md' {PATH}/.gitignore 2>/dev/null \
&& echo "GITIGNORE=already_present" \
|| echo "GITIGNORE=needs_update"
If needs_update:
echo 'Notes/doc-report-*.md' >> {PATH}/.gitignore
git -C {PATH} add .gitignore
git -C {PATH} commit -m "chore: ignore auto-generated doc reports
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>"
Step H — Return result
Print exactly this block with real values filled in:
RESULT:
NAME={NAME}
STATUS=Updated
FILES=N files updated
COMMIT=abc1234
REPORT={PATH}/Notes/doc-report-YYYY-MM-DD.md
ERROR=none
Use STATUS=Skipped if not stale, STATUS=Error if something failed.
Step 3: Collect results and display summary
After all agents return, collect their RESULT blocks.
Sort the rows alphabetically by project name.
Display a markdown table followed by the report file paths:
## Documentation Sweep Complete
| Project | Status | Docs Updated | Commit |
|------------|---------|--------------|---------|
| ProjectOne | Updated | 4 files | mno7890 |
| ProjectTwo | Updated | 8 files | abc1234 |
| ProjectTri | Skipped | — | — |
## Report Files
- /path/to/ProjectOne/Notes/doc-report-YYYY-MM-DD.md
- /path/to/ProjectTwo/Notes/doc-report-YYYY-MM-DD.md
- ... (one per project, sorted alphabetically)
Step 4: Offer to save
Ask the user:
Save a copy of this summary?
[summary] · [list only] · [both] · [no]
- summary — write the full table + report file paths to
~/.claude/doc-sweep-YYYY-MM-DD.md
- list only — write only the report file paths to
~/.claude/doc-sweep-YYYY-MM-DD.md
- both — write summary table + report file paths to
~/.claude/doc-sweep-YYYY-MM-DD.md
- no — done, nothing saved
After saving, confirm with the full path of the saved file.
