# Loop Engineering Handbook - prompt library (for agents)

This directory is a catalog of governed-loop recipes ('cards').

## How to use
- Read `catalog.json` for the machine-readable index (schema: `catalog.schema.json`).
- Each card lives at its `path` and contains a fill-in-the-blanks prompt that reuses
  the canonical six-field loop contract (`../docs/the-loop-contract.md`).
- Pick a card by `category` and `primitive`, copy its Prompt block, fill the brackets,
  and obey its Done-when + Verify before treating the loop as finished.

## Cards (17)
### engineering
- claim-ledger-security (/goal): A scanner dumped a pile of findings and you want only the ones backed by a runnable repro. -> library/loops/engineering/claim-ledger-security.md
- flaky-test-quarantine (/loop): Flaky tests are eroding trust in CI and you want a loop to detect, quarantine, and file them. -> library/loops/engineering/flaky-test-quarantine.md
- overnight-217-review (/loop): You want an agent to review an open-PR queue continuously, without it re-reviewing unchanged PRs forever. -> library/loops/engineering/overnight-217-review.md
- reproduce-before-fix (/goal): A regression needs fixing and you want to forbid touching source until the bug is reliably reproduced. -> library/loops/engineering/reproduce-before-fix.md
- two-loops-one-repo (/loop): You want two agents (e.g. a refactor loop and a docs loop) working the same repo without clobbering each other. -> library/loops/engineering/two-loops-one-repo.md
### operations
- ad-spend-reconciliation (/goal): You need messy invoices reconciled against a bank export to a hard zero-variance gate. -> library/loops/operations/ad-spend-reconciliation.md
- log-anomaly-patrol (/loop): You want a security-camera loop that watches logs for new error signatures and opens an issue once per signature. -> library/loops/operations/log-anomaly-patrol.md
- saas-spend-audit (/loop): You want a recurring audit of SaaS/cloud spend that flags anomalies but can't run up its own bill. -> library/loops/operations/saas-spend-audit.md
### content
- changelog-from-commits (/goal): You want a release changelog drafted only from the actual commit/PR history, with every entry traceable. -> library/loops/content/changelog-from-commits.md
- deliverability-rescue (/goal): An email/landing page must clear a spam/quality score and you want a writer-vs-scorer loop to climb to it. -> library/loops/content/deliverability-rescue.md
- docs-freshness-patrol (/loop): You want a patrol that flags docs whose code examples or referenced APIs have drifted from the codebase. -> library/loops/content/docs-freshness-patrol.md
### research
- claim-verify-research (/goal): You want a research brief where every claim is independently verified before it is allowed to stay. -> library/loops/research/claim-verify-research.md
- competitive-intel-digest (routine): You want a scheduled digest of changes to a watchlist of competitor pages, with each claim linked to its source. -> library/loops/research/competitive-intel-digest.md
- rfp-questionnaire-pack (/goal): You must answer a long questionnaire and every answer has to cite an approved source or be left as a gap. -> library/loops/research/rfp-questionnaire-pack.md
### data
- csv-normalizer (/goal): You have messy CSVs (mixed formats, currencies, dates, encodings) to normalize to a target schema. -> library/loops/data/csv-normalizer.md
- data-quality-gate (/loop): You want a patrol that checks a dataset/table against quality rules and blocks or flags on violations. -> library/loops/data/data-quality-gate.md
- schema-migration-backfill (/goal): You're backfilling a new column/table and need every affected row migrated with a verifiable invariant. -> library/loops/data/schema-migration-backfill.md