NagareNegishi/DJ-App / .claude/skills/plan-verify
plan-verify
VerifiedHarden a rough implementation plan into a verified one β expand each step and attach real proof, promoting confirmed steps to `π verified`. Command-only, run once the product doc is settled and stable: it spends the real verification budget. Drafts a doc only.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 4.0 KB |
SKILL.md
View on GitHub---
name: plan-verify
description: >
Harden a rough implementation plan into a verified one β expand each step and
attach real proof, promoting confirmed steps to `π verified`. Command-only,
run once the product doc is settled and stable: it spends the real
verification budget. Drafts a doc only.
disable-model-invocation: true
---
# Plan Verify
Verify the rough `impl.md` from `plan-impl` step by step against source code and
current external documentation, attaching proof to every claim. Last skill in the
chain: `plan-product` β `plan-impl` β **`plan-verify`**.
## Input and output
- Read `docs/plans/<slug>/impl.md`. If it is missing, stop and tell the user to
run `plan-impl` first.
- The **only** write is back to that same `impl.md`. Read source files and
external docs freely to check claims, but never change code, run git, or post
to the network yourself. (External research goes through `dev-research`, below.)
## The verification mark
Every step ends at one of:
```
π verified β src: path/to/file.ts:42 internal: the code exists / behaves as claimed
π verified β doc: <url> Β§exact-section external: the tool/approach is current and standard
β unverified (not checked) could not confirm this pass
β unverified (net-new) new code, so there is legitimately nothing to link
```
**No step is marked `π verified` without proof you actually saw this session.**
- **Internal claims** β open the file and cite `file:line` you read. A step that
says "add a handler in `FooService`" is verified only after you open the file
and confirm `FooService` exists at that line. If it does not exist and the step
is building it, that is `β unverified (net-new)` β expected, not a failure.
- **External claims** β a library API, a "standard" approach, a tool that may
have changed since the model's training cutoff. Verify these by invoking the
`dev-research` skill, which checks against current official docs before
asserting. Cite the exact doc page and section it returns. Do not verify an
external claim from memory β the cutoff makes that a real hallucination risk.
Leave a step `β unverified (not checked)` when you genuinely could not confirm
it this pass (docs unreachable, ambiguous, out of scope). Never downgrade an
honest unknown into a fake `π verified`.
## Maturity is a separate axis
The maturity mark (`π±`/`π€`/`π€`/`β
`) is independent of verification. Advance a
step to `π€ ai-audited(<model>)` on your own. Stamp `π€ human-ok` or `β
settled`
only when the user explicitly instructs you to set that mark on a named step:
ask, wait for the instruction, then stamp β one step per instruction. Never stamp
either mark on your own initiative, and never reuse one step's instruction for
another. A step can be `π verified` against source while still `π€ ai-audited` on
maturity; that is a valid, expected combination.
## Workflow
1. Read `impl.md`. Work step by step, expanding each rough step into concrete
detail as you go.
2. For each claim in a step, decide internal or external, then verify:
- Internal β open the file, confirm, cite `src: file:line`.
- External β invoke `dev-research`, cite `doc: <url> Β§section` it returns.
- New code with nothing to check β `β unverified (net-new)`.
- Couldn't confirm β `β unverified (not checked)`, and say why in the step.
3. Stack both marks on the step's mark line, e.g.
`π€ ai-audited(opus-4.8) Β· π verified β src: backend/export.ts:42`.
4. Update the `## Maturity` header in the same pass so it can't drift.
## human-writing
Follow the `human-writing` skill for any expanded prose.
## After writing
Do not render the doc in chat. Show the path and report a one-line tally: how
many steps are now `π verified` (src vs doc), how many `net-new`, and how many
still `not checked` and why. Flag anything the verification pass proved wrong in
the product plan so the user can revisit it.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |