Self review

Skill Firework-fufu/personal-agent-starter-kit/skills/self-review

Pressure-test important plans using a challenger pass and a coverage pass before the main agent makes a final judgment.From its SKILL.md

Install

npx -y skills add Firework-fufu/personal-agent-starter-kit --skill self-review

Assembled 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.

SKILL.md

3.5 KB, 735 tokens by cl100k_base, as published. Nobody here has run it

Self Review

Use this for important decisions. It is not a normal checklist. It separates plan creation, challenge, coverage checking, and final judgment.

Triggers

Use this skill when:

the user asks for a serious review;
the plan affects project direction, architecture, business strategy, or public release;
the agent may be too attached to its own proposal;
the user asks for a red-team or adversarial check.

Do not use it for small edits or routine tasks.

Input Format

Use a JSON task file:

{
  "context": "What problem are we solving?",
  "constraints": "What must not be violated?",
  "plan": "What proposal should be reviewed?"
}

context and plan are required. constraints is optional but strongly recommended.

Roles

Role	Job	Not Allowed
Plan author	Writes the plan in the main conversation	Does not review itself
Challenger	Finds risks, edge cases, and failure paths	Does not comfort or propose alternatives
Coverage checker	Checks whether the plan already addresses each challenge	Does not rank or fix the plan
Final judge	Main agent decides what to change or accept	Does not ignore uncovered risks

Steps

Step 1: Prepare The Plan

Write the plan clearly enough for another model to review.
Include rejected options and known risks when available.
Avoid private context that should not be sent to an API.

Checkpoint: the plan can be understood without reading the current chat.

Step 2: Write Task JSON

Use examples/decision-review-task.json as a template.

Checkpoint: the JSON parses and contains context, constraints, and plan.

Step 3: Run Offline Self-Test First

python3 scripts/self_review.py self-test

Checkpoint: the command exits successfully.

Step 4: Run Review

If you configured ~/.config/personal-agent.env, run:

python3 scripts/self_review.py review examples/decision-review-task.json --challenge-out tmp/challenge.json --coverage-out tmp/coverage.json --out tmp/review.json

On Windows, use python if python3 is not available.

The script uses Python standard library HTTP calls, so it does not require third-party Python packages.

Checkpoint: challenge and coverage files are generated.

Step 5: Final Judgment

The main agent must summarize results like this:

| Challenge | Coverage | Risk Level | Decision |
|---|---|---|---|
| ... | covered/partial/uncovered | critical/important/acceptable | change now / defer / accept |

Checkpoint: every important challenge has a decision.

Step 6: Feedback

If the review changes the plan, record the decision in memory/decisions.md or feedback/log.md.

Checkpoint: future agents can see why the plan changed.

Failure Handling

Missing API key: run self-test only, or configure OPENAI_API_KEY.
Missing model: set OPENAI_MODEL or pass --model.
Bad JSON: fix the task file before calling the API.
Challenger proposes solutions: treat as role drift and rerun.
Coverage checker gives advice: treat as role drift and rerun.
Many uncovered risks: do not automatically reject the plan; classify risks first.

Version History

v1 (2026-05-03): Initial public starter-kit version.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most review quality skills give in 735 tokens

Counted across 1,048 of the 1,783 authors here whose files we hold, read 2026-08-07

Ask questions one at a timein 81 of 1048, across 64 files
Provide a recommended answer for each questionin 73 of 1048, across 50 files
Explore the codebase instead of asking answerable questionsin 66 of 1048, across 42 files
Resolve dependencies between decisions one-by-onein 42 of 1048, across 17 files
Interview the user relentlessly about the planin 38 of 1048, across 13 files
Order findings by severityin 31 of 1048
Resolve each branch of the decision treein 27 of 1048, across 5 files
Run a grilling sessionin 26 of 1048, across 5 files
Update CONTEXT.md immediately when a term is resolvedin 26 of 1048, across 11 files
Propose precise canonical terms for vague languagein 25 of 1048, across 7 files
Create documentation files lazilyin 24 of 1048, across 5 files
Assign severity to every findingin 24 of 1048

Said here and by no other author read

separate plan creation, challenge, coverage, and judgment
write the plan clearly for external review
avoid sending private context to an API
run the review using the task JSON
summarize every important challenge with a decision
record changed plans in a feedback log

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.