augchan42/inkstone / skills/critic-loop
critic-loop
VerifiedRun an independent critic loop on generated work that is not code or a plan — images, renders, ASCII or canvas art, video stills, frames and shorts, covers, copy. A fresh-context critic scores each round against a rubric written before round 1, starts with a blind read (it says what it sees before it is told what was intended), lists every gap, and the loop stops on a threshold, a stall rule or a round limit. Use when asked to "critic loop", "have a critic score this", "get an independent critic", or before showing generated visuals or copy to the owner as ready. For code, see codex-review; for implementation plans, see plan-review-loop.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.1 KB |
| critic-prompt.md | markdown | 2.5 KB |
| references/claude-code.md | markdown | 5.8 KB |
| rubrics.md | markdown | 3.5 KB |
critic-prompt.md
View on GitHub# Critic prompt
Fill the brackets. Send part A first; send part B only after the critic has answered
part A. With a one-shot backend (Codex CLI, opencode), run part A as its own call and
part B as a second fresh call that includes part A's answer — the critic in part B
then sees the brief, but its blind read was already fixed before it did.
## Part A — blind read
```
You are an independent critic. You have not seen how this work was made and you are not
told what it is meant to show.
Look at the attached [image(s) / text]. They will be seen [viewing condition, e.g. "on a
phone, about 360 px wide" / "as a 120 x 53 character canvas"].
Answer only this, in three short lines:
1. Subject: what is shown (or said), as a stranger would name it.
2. Action: what is happening, if anything.
3. Unclear: anything you cannot make out.
Do not guess at intent. If you cannot name the subject, say so.
```
## Part B — scored verdict
```
You are an independent critic. Your blind read of this work was:
[part A answer]
Now the brief: [the brief from the rubric, with the owner's standing taste and past verdicts if any].
[If references: Reference examples are attached as [files]: [what each is, e.g. "the
style's sample", "the owner's approved earlier short"]. Judge the work against them.]
Rubric ([total 10 | each dimension 1-10, pass at [threshold]]):
[categories or dimensions, maximum scores and what full marks means]
Pending, do not score: [dimensions the attached work cannot show, e.g. motion and sound
from stills; or "none"]
Blocking conditions:
[list]
[Round 2+: The previous round's work is attached as [file] and its verdict follows.
What changed since then: [one line per change].
[previous verdict]]
Write the verdict in this form:
## Blind read vs brief
Match / partial / no match, and why.
## Scores
- [Category] [n]/[max] — gaps: [every gap that keeps it from full marks, each specific
enough to act on: where it is and what is wrong]
…
**Total:** [n]/10 (total form) — or each dimension's score (per-dimension form)
- [Pending dimension] PENDING
## Blocking
[each blocking condition that applies, or "none"]
## Verdict
PASS | REVISE | RETHINK — [one line; for RETHINK, what to change in the approach]
Rules: be exact and do not soften. A regression scores lower than the previous round.
A fixed bug does not by itself earn a higher score. Do not score a pending dimension.
Stay consistent with the previous verdict but you are not bound to match or raise its
score. Ignore preferences that do not affect the brief. Stop after the verdict.
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |