yogsoth-ai Spec Self-Review
You are auditing a Research Spec for quality before presenting it to the user. This review is MANDATORY and cannot be skipped.
Checklist
Run each check. If any fails, fix the issue inline immediately.
1. Placeholder Scan
Search the spec for: "TBD", "TODO", "...", "fill in", "to be determined", or any vague language where specifics are needed.
- Every field must have concrete content
- "Adequate coverage" is NOT acceptable — use numbers
2. Internal Consistency
- Do Stage numbers match in backtrack conditions? (e.g., "→ Stage 3" actually exists)
- Do "Expected Input" references point to outputs that prior Stages actually produce?
- Are campaign/strategy names spelled consistently throughout?
3. Scope Check
- Is the spec executable in 3-10 sessions?
- If >10 stages, suggest splitting into sub-specs
- If <3 stages, verify this isn't too shallow for the North Star
4. Ambiguity Check
- Could any completion criterion be interpreted two different ways?
- Are "Focus Areas" specific enough that two different CCs would focus on the same things?
- If ambiguous, pick one interpretation and make it explicit
5. Context Protocol Check
- Does EVERY Stage have: context-init step + at least one context-checkpoint step + campaign-end checkpoint?
- Are topic-slugs in context-init descriptive and unique?
6. Quantification Check
- Are ALL completion criteria numeric or objectively verifiable?
- "≥20 papers" is good. "Sufficient papers" is not.
- "≥3 validated gaps with evidence" is good. "Identify gaps" is not.
Output
If all checks pass: report "Spec self-review: PASS" and proceed. If any check fails: fix inline, then report what was fixed.