plan-critic
/plan-critic
Use when a plan, meaning a list of tasks with sizes, dependencies and verify lines, must be checked before it is built. Runs a deterministic script first, then a 0 to 2 rubric, and returns PASS, REVISE or BLOCK.
- Author
- KaaS-team ↗
- Repository
- blueprint ↗
- Licence
- Apache-2.0
- Last repository change
- 12.11.2026
Assessment and security
- Files in this package
- 8
- Scripts
- 1
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 12.11.2026, without references to retired models or APIs.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
- 1 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/check-plan.mjs
What it does and when to use it
Use when a plan, meaning a list of tasks with sizes, dependencies and verify lines, must be checked before it is built. Runs a deterministic script first, then a 0 to 2 rubric, and returns PASS, REVISE or BLOCK.
Original author description
Use when a plan, meaning a list of tasks with sizes, dependencies and verify lines, must be checked before it is built. Runs a deterministic script first, then a 0 to 2 rubric, and returns PASS, REVISE or BLOCK.
How it works
- Plan format
- Run the script
- Rubric
- Verdict
SKILL.md preview
---
name: plan-critic
description: Use when a plan, meaning a list of tasks with sizes, dependencies and verify lines, must be checked before it is built. Runs a deterministic script first, then a 0 to 2 rubric, and returns PASS, REVISE or BLOCK.
---
# plan-critic
This is the one plan rubric. The factory's `factory-plan-rubric` points here, and
this repo carries no second plan rubric skill.
Checks a plan in two steps. The script finds structural faults. The rubric
judges the plan's content. Run them in this order: the rubric is only worth
scoring once the script passes.
## Plan format
A plan is a Markdown file with one heading per task, then `- key: value` lines:
```
## T1: Add the retry helper
- size: S
- depends: none
- files: src/retry.mjs (new), src/upload.mjs
- verify: a unit test shows a failed upload is retried twice
```
- `## T<n>: <title>` starts a task. Ids are `T1`, `T2`, and so on.
- `size` is `S`, `M` or `L`.
- `depends` is a comma-separated list of task ids, or `none`.
- `files` is a comma-separated list. Each path must exist relative to the
working directory, or end with ` (new)` when the task creates it.
- `verify` names the test or acceptance criterion that proves the task.
## Run the script
```
node plugins/blueprint-review/skills/plan-critic/scripts/check-plan.mjs PLAN.md
```
Run it from the repo root, because `files` paths resolve against the working
directory. Other repos should point the command at the script's own path. It prints one JSON
object, `{ok, findings[]}`, and exits 0 when `ok` is true, 1 when it is false.
Each finding names its `task` (or `null` for a plan-wide line) and its `rule`:
| rule | fires when |
| --- | --- |
| `no-tasks` | the plan has no `## T<n>:` task headings |
| `no-verify` | a task has no `verify` line |
| `no-size` | a task has no size of S, M or L |
| `unknown-dependency` | a task depends on a task id that does not exist |
| `cycle` | a task sits on a dependency cycle |
| `unknown-file` | a file is neither present nor marked `(new)` |
| `placeholder` | the plan contains `TBD`, `TODO`, `???` or `FIXME` |
If `ok` is false, fix every finding and run the script again before scoring.
## Rubric
Score each item 0, 1 or 2. Read the whole plan before you score.
1. **intent**: the plan covers what the request asked for.
2. **scope**: the plan stays inside the request. Work the request did not ask
for scores 0.
3. **verification**: every task says how it is verified, not only some.
4. **risk**: the plan names its risk, such as data, security or a public API.
5. **estimate**: every task has a size, and the plan has a total.
Write the scores as a JSON object keyed by the five item names, for example
`{"intent": 2, "scope": 2, "verification": 2, "risk": 1, "estimate": 2}`.
A missing score, or one that is not an integer from 0 to 2, counts as 0.
## Verdict
Pass the script's findings and the scores to `verdict({findings, scores, round})`,
exported by `scripts/check-plan.mjs`, or apply the same rule by hand:
- **PASS** needs no script findings, every item at 1 or more, and a total of 8 or more.
- **REVISE** is any other result on round 1. Send the findings and the low items
back to the plan's author, who revises once.
- **BLOCK** is any other result on round 2. Stop and report the plan as blocked.
There is at most one revise round. A second failure is BLOCK.
Files in this package (8)
- SKILL.md 3 KB
- fixtures/clean.md 1 KB
- fixtures/clean.scores.json 1 KB
- fixtures/cycle.md 1 KB
- fixtures/no-verification.md 1 KB
- fixtures/scope-creep.md 1 KB
- fixtures/scope-creep.scores.json 1 KB
- scripts/check-plan.mjs 5 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · KaaS-team/blueprint | 100 | Apache-2.0 |
| cunhaax/ai-workflow | 100 | MIT |
| smykla-skalski/sai | 100 | MIT |
Database snapshot: 13.11.2026 · package version 4365ab7a861c (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100code-critic
Reviews code changes against project standards and the approved plan; returns PASS/FAIL/NEEDS_DECISION per checklist item plus open questions. Use when the user asks for a code review of a branch or change, or after implementation and before opening a PR. - 100plan-validate
Deterministic plan.json validation via Python script, replacing most plan-validator agent calls. Performs JSON parsing, schema validation, task checkbox regex, required section checks, sync validation, and data extraction. Only semantic consistency checks (storage/query… - 100visual-verdict
Screenshot comparison QA for frontend development. Takes a screenshot of the current implementation, scores it across multiple visual dimensions, and returns a structured PASS/REVISE/FAIL verdict with concrete fixes. Use when implementing UI from a design reference or verifying visual correctness. - 100review-plan
Use when a Plan is complete and ready for review. Two-stage review -- spec contract compliance first, code quality second. Block on contract failure before reviewing code. - 100writing-plans
Use when the user requests an implementation plan or multi-step work needs documented task boundaries, dependencies, and acceptance checks. - 100codex-plan-writer
Use for complex coding requests that need a structured implementation plan with dependencies, checkpoints, and verification criteria.