Skip to content

RECIPES.md

Latest commit

 

History

History
187 lines (132 loc) · 4.85 KB

RECIPES.md

File metadata and controls

187 lines (132 loc) · 4.85 KB
Raw

Local Automation And Agent Archive Recipes

Copy-paste dry-run recipes for scheduled regeneration, prompt-library refresh, post-run summaries, and common agent archive jobs. Every command below was verified locally without provider spend unless noted.

Related docs:

Setup Once

cd /path/to/rafiki
cp config/scheduled-regen.json.example config/scheduled-regen.json
# edit jobs in the gitignored local config

Keep private or machine-specific paths in config/scheduled-regen.json. The tracked example uses repo-relative paths only.

Scheduled Regeneration

Goal: Preview which evergreen prompt batches are due, then run only due jobs.

python generate.py regen --dry-run
python generate.py regen --yes

Use the tracked example config when the local config does not exist yet:

python generate.py regen --config config/scheduled-regen.json.example --dry-run

Expected output: A table of configured jobs with [DUE] or skip status, interval days, and last-run age. A real --yes run creates fresh run-*/ directories under each job's output_dir and does not auto-approve images.

Prompt Library Refresh

Goal: Force one evergreen batch to regenerate after model or style changes, then rebuild review surfaces.

python generate.py regen --config config/scheduled-regen.json.example --dry-run
python generate.py regen --name upgrade-newsletter-heroes --yes
python generate.py view upgrade-newsletter --all-runs
python generate.py library

Replace the job name with any configured name field. Expected output: A new run-*/ directory, updated comparison viewer HTML, and a rebuilt output/library.html.

Post-Run Summary

Goal: Run due jobs, refresh the master library, and report paths for human or agent review.

python generate.py regen --dry-run
python generate.py regen --yes
python generate.py library
python generate.py archive-health --json

Expected output: Due-job execution log, output/library.html, and a JSON health summary with project/run counts. Report generated run-*/ paths only; do not send private prompt text or provider keys externally. Treat health against the live local archive as advisory; use a disposable or fixture --output-dir for release checks.

Agent Recipe: Find Unused Good Assets

Goal: Search the registry for approved or starred candidates that have not been exported.

CLI:

python generate.py registry search "workshop" --limit 20
python generate.py registry export --format json --dry-run

MCP (rafiki_registry_search):

{"query": "workshop", "limit": 20}

Expected output: Matching registry entries with titles, tags, approval state, and export metadata when indexed.

Agent Recipe: Compare These Runs

Goal: Inspect archive health and rebuild viewers for side-by-side review without regenerating images.

CLI:

python generate.py archive-health --cleanup-report
python generate.py view <project> --all-runs
python generate.py viewer-rebuild <project> --all-runs --dry-run

MCP:

{"output_root": "/absolute/path/to/output", "cleanup_report": true}
{"project": "<project>", "all_runs": true, "dry_run": true}

Expected output: Health report with duplicate filenames and cleanup candidates; dry-run viewer rebuild paths for each run.

Agent Recipe: Prepare A Web Gallery Shortlist

Goal: Export approved or latest-run assets and rebuild the master library for a static review handoff.

CLI:

python generate.py registry index
python generate.py registry export --format json
python generate.py canva-export <project> --dry-run
python generate.py library

MCP:

{"format": "json", "dry_run": true}
{"project": "<project>", "dry_run": true}

Expected output: Registry export preview, Canva bundle path preview, and output/library.html for local browsing.

Agent Recipe: Audit Stale Generated Assets

Goal: Read-only audit of missing files, malformed manifests, sidecar orphans, and conservative cleanup candidates.

CLI:

python generate.py archive-health --json
python generate.py archive-health --cleanup-report

For release verification, run the same commands against a controlled fixture or disposable output root with --output-dir.

MCP (rafiki_archive_health):

{"cleanup_report": true}

Expected output: JSON with ok, summary counts, missing-image lists, and suggested dry-run cleanup commands. No files are mutated.

Verification

After editing these recipes:

npm run docs:check
npm run smoke:dry-run

For scheduled regen specifically, always dry-run before --yes when adding or changing jobs.