Iterative self review
A collection of practical, production-ready AI agent skills for the skills.sh ecosystem.
npx -y skills add BlizzardBlast/frey-skills --skill iterative-self-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use only when explicitly invoked for an issue ledger, remediation request, repeated fix-and-recheck loop, or post-review repair. Accepts code-review findings or user-specified issues, fixes them in bounded passes, compares against a resolved task/default-branch baseline, and reports scoped verification without claiming the whole repository is clean.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
4.2 KB, 818 tokens by cl100k_base, as published. Nobody here has run it
Iterative Self-Review
Activation boundary
Use this skill only when the user explicitly requests iterative self-review, repeated remediation, or fixes for a provided issue ledger. Do not invoke it implicitly for ordinary implementation or one-off review.
Inputs may be:
- A
code-reviewfinding ledger. - Explicit user remediation instructions.
- A scoped list of failing tests, review comments, or defects to fix.
Default pass budget
Run at most 3 review/fix/verify passes by default. Stop early only when the scoped issues are resolved and the required verification for those issues has passed or is honestly marked blocked. After pass 3, stop and ask for direction if any in-scope item remains unresolved or blocked. Do not run pass 4 or any later pass unless the user explicitly asks for additional passes after seeing the pass-limit status; never continue past the limit silently.
Do not claim “zero issues”, “all clean”, or whole-repository correctness. Report only scoped resolution and verification evidence.
Baseline setup
Before editing, record a working-state baseline:
- Branch name and dirty/untracked status.
- Files in scope and incoming issue IDs.
- Existing failing or skipped verification that may affect claims.
Resolve the comparison base in this order:
- User-provided base.
origin/HEAD.origin/main.main.origin/master.master.
If no base can prove broader regression claims, set BASELINE_LIMITED and state what comparison remains valid, such as current working tree only or issue-specific files only.
Required loop
For each pass:
- Review the unresolved ledger items and current code against the baseline.
- Plan the smallest safe fix set for this pass.
- Edit only in-scope files needed to resolve the ledger.
- Verify with the most relevant focused commands or evidence.
- Update the ledger:
resolved,unresolved,blocked, ordeferred by user scope. - Stop if scoped issues are resolved; otherwise continue until the pass budget is exhausted.
Fix order is P0 -> P1 -> P2 -> P3 unless the user narrows scope. If a request is P1-only, do not opportunistically fix P2/P3 items except for necessary supporting edits.
Conflict and blocker rules
- If two fixes conflict, choose the safer requirement-preserving path and document the tradeoff.
- If the same issue toggles across passes, stop rather than churn.
- If verification cannot run or required context is missing, mark the affected item
blockedwith exact evidence. - If the 3-pass default limit is reached with unresolved items, stop, report what remains, and ask for direction. Continue to pass 4 or any later pass only after the user explicitly requests additional passes after that status is reported.
References
Load only as needed:
references/baseline-and-pass-rules.mdfor baseline commands, pass ledger states, andBASELINE_LIMITEDwording.references/issue-ledger-format.mdfor a compact remediation ledger template.references/evaluation-playbook.mdonly when evaluating this skill.
Output format
Use this concise structure:
Baseline- base used, working-state notes, and
BASELINE_LIMITEDif applicable.
- base used, working-state notes, and
Pass N- issue IDs attempted, edits made, verification run, and ledger updates.
Final statusRESOLVED,PARTIAL, orBLOCKEDfor the requested scope.
Remaining concerns- unresolved blockers, skipped out-of-scope items, and any verification limits.
Completion conditions
This skill is complete when one of these is true:
- All in-scope ledger items are resolved and required verification is recorded.
- The pass budget is exhausted and remaining work is reported.
- A blocker prevents further safe progress and is reported with evidence.