Validate the structural integrity of a recipe after it has been created or modified in the cookbook file
Validates the structural integrity of a recipe after it has been created or modified in the cookbook file. Catches errors before they get committed — missing fields, naming violations, duplicate IDs, infrastructure count drift, and more. Run this automatically after every recipe creation or modification workflow. Use full mode for new recipes and structural changes. Use quick mode for minor wording edits. Always address FIX_REQUIRED issues before committing the cookbook.
Cowork Recipe Integrity Validator
Tags: Introduced in Beta, Validation, Integrity, QA, Checklist, Fernand, MPCS, Cookbook, Cowork
TL;DR
What It Does
Validates the structural integrity of a recipe after it has been created or modified in the cookbook file. Catches errors before they get committed — missing fields, naming violations, duplicate IDs, infrastructure count drift, and more.
How It Works
The recipe loads the target recipe from the cookbook file and runs a Fernand-driven 10-point checklist. Checks A through F cover structural fundamentals: required fields, recipe ID format, version string format, category validity, dependency existence, and step numbering. Checks G through J cover deeper integrity: description uniqueness, CRAFT status markers, no duplicate IDs, and infrastructure count accuracy. Quick mode runs only A–F. The result is a RECIPE_REVIEW_REPORT with an ACCEPT or FIX_REQUIRED verdict.
What To Expect
A structured report showing each check’s pass/fail status. If all checks pass, you get an ACCEPT verdict and the recipe is clear to commit. If any check fails, you get FIX_REQUIRED with specific details on what to fix. Fernand produces the report; Cat handles remediation if issues are found.
Best Results When You
Run this automatically after every recipe creation or modification workflow. Use full mode for new recipes and structural changes. Use quick mode for minor wording edits. Always address FIX_REQUIRED issues before committing the cookbook.
Typical Time
Under 1 minute for quick mode (6 checks). 1–2 minutes for full mode (10 checks with cross-reference analysis).
Difficulty
Medium — the validator itself is automatic, but understanding the check results and performing remediation for FIX_REQUIRED verdicts requires familiarity with CRAFT recipe structure.
How To Start
This recipe is typically called automatically by other recipes after creating or modifying a cookbook entry. You can also invoke it directly.
STEP 1Specify the Recipe to Validate
Provide the recipe ID of the recently created or modified recipe and choose the validation depth.
Available parameters
-
recipe_id_to_validate
· string · required
The recipe ID that was just created or modified. Focuses validation on that specific recipe while also running cookbook-wide integrity checks. -
validation_mode
· string · optional · default: full
Validation depth. Options: “full” (all 10 checks plus cross-references), “quick” (checks A–F only — structural, skip cross-reference and infrastructure checks). -
called_by
· string · optional
Recipe ID of the calling recipe, used for audit logging.
Invocations
Full Validation (Default)
#H->AI::Directive: (Validate recipe integrity post-creation/modification with recipe_id_to_validate=“RCP-CWK-050”)
Quick Validation
#H->AI::Directive: (Validate recipe integrity post-creation/modification with recipe_id_to_validate=“RCP-CWK-050”, validation_mode=“quick”)
Called by Another Recipe
#H->AI::Directive: (Validate recipe integrity post-creation/modification with recipe_id_to_validate=“RCP-CWK-050”, called_by=“RCP-CWK-ADM-077”)
Usage Examples
Three scenarios — full validation passing, full validation with failures, and quick validation after a minor edit.
Example 1 · Full validation passes
User:
Validate the recipe I just added: RCP-CWK-050.
Fernand:
Executing recipe integrity checklist. Check A — Required Fields: PASS. Check B — Recipe ID Format: PASS. [Checks C–J: PASS.] RECIPE INTEGRITY VALIDATION: ACCEPT. 10/10 checks passed — recipe is structurally sound.
Example 2 · Full validation finds issues
User:
Validate RCP-CWK-051 after my edits.
Fernand:
Checks A–H: PASS. Check I: FAIL — found 4 occurrences, possible duplicate. Check J: FAIL — declared 48 but found 49. FIX_REQUIRED. 8/10 passed, 2 failed.
Cat:
Fernand flagged 2 issues. Investigating duplicate ID and count drift.
Example 3 · Quick validation after minor wording change
User:
I just fixed a typo in CWK-030’s description. Quick validate.
Fernand:
Quick validation (6 checks). A: PASS. B: PASS. C: PASS. D: PASS. E: PASS. F: PASS. ACCEPT. 6/6 passed.
How AI Reads This Recipe
When this recipe is triggered, the AI validates a recipe’s structural integrity through a 4-step pipeline with Fernand as the validator persona. The AI should:
- LOAD THE RECIPE under review by reading the cookbook file and locating the target recipe via Grep. Parse the full Recipe() object. If the recipe is not found, produce a FIX_REQUIRED report immediately.
- EXECUTE THE 10-POINT CHECKLIST (Fernand persona). In full mode, run all 10 checks; in quick mode, run A–F only. Checks: A — Required Fields present. B — Recipe ID matches RCP-CWK-NNN-NAME-vN.NNa. C — Version string matches vN.NNa. D — Category is valid (SESSION, SAFETY, BRIDGE, or QA). E — Dependencies exist in cookbook or are Core references. F — Step numbering is sequential. G — Description not copy-pasted from parent. H — Required CRAFT status markers present. I — No duplicate recipe IDs. J — Infrastructure counts match actual.
- PRODUCE RECIPE_REVIEW_REPORT with passed/failed counts, issue details for each failure, and an ACCEPT or FIX_REQUIRED verdict.
- REPORT AND HAND OFF. If ACCEPT, report success. If FIX_REQUIRED, return control to Cat for remediation with specific failure details. Log via audit trail (R-12).
This recipe implements R-2 (QA Phase C) and is driven by the Fernand persona via MPCS routing. Designed to be called automatically after recipe creation or modification workflows. The audit trail logs every validation run. Error handling covers cookbook not found and Recipe() parse failures.
When to Use This Recipe
Use this recipe when you:
- Have just created a new recipe in the cookbook and need to verify structural integrity before committing.
- Have modified an existing recipe and want to confirm no structural damage was introduced.
- Want to check infrastructure count accuracy after multiple recipe additions or deletions.
- Are building a recipe creation workflow and need a post-creation validation gate.
Do not use this recipe when:
You are reviewing recipe CONTENT quality (does the recipe do the right thing, are the steps well-designed). CWK-023 checks STRUCTURE, not content. For content review, use the evaluation step in the CRPW pipeline or the CWK-ADM-084 retrospective.
Recipe FAQ
Q.What are the 10 checks?
A: Required Fields present. B: Recipe ID matches naming convention. C: Version string matches vN.NNa format. D: Category is valid. E: All dependency references exist. F: Step numbering is sequential. G: Description not copy-pasted from parent. H: Required CRAFT status markers present. I: No duplicate recipe IDs. J: Infrastructure counts match actual.
Q.What is quick mode?
Quick mode runs only checks A through F — the structural fundamentals. It skips cross-reference checks (G, H) and infrastructure checks (I, J). Use it for minor edits like wording changes that do not affect recipe structure.
Q.Who is Fernand?
Fernand is the QA validator persona in the MPCS (Multi-Persona Communication System). He executes the checklist with a critical eye, producing pass/fail results for each check. Cat handles remediation if Fernand finds issues.
Q.What does FIX_REQUIRED mean?
At least one check failed. The report lists every failed check with specific details (e.g., “Missing: prompt_template” or “DRIFT: Declared 48 but found 49 recipes”). Cat must address each failure before the cookbook can be committed.
Q.What are valid Cowork categories?
Four: CAT-CWK-SESSION (session lifecycle recipes), CAT-CWK-SAFETY (safety and guardrail recipes), CAT-CWK-BRIDGE (Core extension recipes), and CAT-CWK-QA (quality assurance recipes).
Q.How does the audit trail work?
Every validation run is logged with: the recipe validated, the calling recipe (if invoked by another recipe), the validation mode, checks passed/failed, and the verdict. This implements the R-12 audit trail requirement from QA Phase C.
Version History
Changes to this recipe over time. Most recent first.
v1.00a
2026-02-28
Initial release. Fernand-driven post-recipe validation: 10-point checklist (required fields, ID format, version string, category validity, dependency existence, step numbering, keyword uniqueness, status markers, duplicate ID detection, infrastructure count verification), 2 validation modes (full, quick), RECIPE_REVIEW_REPORT with ACCEPT/FIX_REQUIRED verdicts, MPCS routing, audit trail logging (R-12), error handling for missing cookbook and parse failures.
Get this recipe with CRAFT for Claude Cowork
Cowork recipes ship bundled with CRAFT for Claude Cowork — there’s no separate download. Clone the framework once, and your AI runs every recipe automatically when invoked.
Pull anytime to stay on the latest version — free to clone, no login or email required.
Then start your session
Once CRAFT is in your project folder, open a new Cowork session and ask Claude to initialize. For example:
You
Please initialize my CRAFT session.
Claude
CRAFT session ready. Your project is loaded, your persona is active, and your recipes are available. What would you like to work on?
Not familiar with Git? Download as a ZIP
No command line needed. Just download, move, and unzip:
- Open the CRAFT framework repo on GitHub.
- Click the green Code button, then choose Download ZIP.
- Move the downloaded ZIP into your Claude Cowork project folder.
- Unzip it: double-click on Mac, or right-click → Extract All on Windows.
