# Self Critique Protocol > Self Critique Protocol - Skill: `handsomeboy990/self-critique-protocol` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add handsomeboy990/self-critique-protocol` - Raw SKILL.md: https://api.skillmd.com/api/skills/handsomeboy990/self-critique-protocol/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: Handsomeboy990 (https://skillmd.com/u/handsomeboy990) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/handsomeboy990/self-critique-protocol --- # Self Critique Protocol Mandatory self-assessment protocol. No text produced by a skill in this tree is delivered without passing through it. It is not an opinion; it is a reproducible procedure. This is the creative writing depth. `shared/self-critique` selects perspectives across every domain and delegates here for fiction, poetry and screenplay, because the axes and thresholds below are more demanding than anything a general panel would produce. ## 1. Absolute rule Generate, assess, correct, reassess. A text not reassessed after correction is not finished. The protocol runs at least once, and repeats until the threshold is met, up to three cycles. On the third failure, the text is rewritten from the scene sheet rather than patched. ## 2. The eleven axes Each axis is scored 0 to 5. The score must be justified by evidence taken from the text and quoted. A score with no evidence is void and counts as 0. ### 1. Narrative quality Does the scene tell something, or only describe? Is there movement, progression, transformation? ### 2. Consistency Internal contradictions, against the bible, against earlier chapters, against the continuity register. ### 3. Rhythm Alternation of sentence and paragraph lengths, match between density and tension, presence of unintended dead time. ### 4. Characters Does each want something? Do they act according to their sheet? Do they have their own voice? Are they allowed to exist outside the plot? ### 5. Dialogue Typographic conformity, subtext, differentiation, absence of exposition, economy of incises. ### 6. Emotion Are the constitution's four supports present: stake, resistance, precise physical manifestation, irreversible consequence. ### 7. Originality Is the treatment the one anyone would have written? Is there at least one choice nobody expected? ### 8. Credibility Can the reader believe the facts, the reactions, the durations, the claimed competences? ### 9. Repetition Words, images, sentence structures, recurring gestures, across a three hundred word window. ### 10. Cliche Worn formulas, stock situations, unsubverted stock characters. ### 11. Logic Causal chaining, plausible decisions, absence of convenience and favourable coincidence. ## 3. Scale | Score | Meaning | |---|---| | 0 | absent, or contrary to the constitution | | 1 | seriously deficient | | 2 | insufficient, correction mandatory | | 3 | acceptable, publishable without pride | | 4 | good, at professional level | | 5 | remarkable, stands alone out of context | Delivery threshold: no axis below 3, average at or above 3.8. For an opening or a turning-point chapter, the threshold rises to 4.2. ## 4. Procedure ### Pass 1: cold read Read the text with no intention of correcting, noting only where attention slips. Mark them; correct nothing. ### Pass 2: scoring Fill the grid, one axis after another, quoting one piece of evidence per score. Scoring two axes at once is forbidden: each axis is its own read. ### Pass 3: diagnosis For every axis under 4, write the cause, not the symptom. For example, `the dialogue is flat` is a symptom; `both characters want the same thing` is a cause. ### Pass 4: correction Correct in decreasing order of severity. One correction of a cause beats ten surface corrections. Never fix one axis by degrading another: verify after each major correction. ### Pass 5: reassessment Rescore the axes touched. If the threshold is met, deliver with the grid. If an axis stays under 3 after three cycles, rewrite. ## 5. Unblocking questions When an axis stalls, apply the matching question: | Axis | Unblocking question | |---|---| | Narrative | What does the character lose in this scene? | | Consistency | Who knows what, and since when? | | Rhythm | Which paragraph can I delete and lose nothing? | | Characters | What does the one who speaks least want? | | Dialogue | What are they refusing to say? | | Emotion | What gesture would replace the named emotion? | | Originality | What was the second idea I had? | | Credibility | Would a practitioner laugh reading this? | | Repetition | Which word appears three times on one page? | | Cliche | Have I read this sentence somewhere else? | | Logic | Why does he not do the simplest thing? | ## 6. Output format The protocol always produces three things: 1. the scored grid, with evidence; 2. the ordered list of corrections applied; 3. the revised version. No partial delivery. ## 7. Anti-complacency Three rules against self-validation: - Scoring 5 more than once per grid requires exceptional evidence. - Any grid averaging above 4.5 on the first cycle is suspect and is rechecked by `quality/literary-critic`. - The protocol looks for what is wrong. It does not write praise. ## 8. Auto-critique of the protocol The protocol applies to itself. After each use, check: - is every score backed by a quotation from the text; - does the diagnosis reach a cause, or stop at the symptom; - did a correction degrade another axis without being detected; - does the first cycle average exceed 4.5, which signals complacency; - did the number of cycles stay at three or fewer. A protocol run with no quoted evidence is void and is repeated. ## 9. Interfaces - Upstream: every production skill. - Downstream: `story-doctor`, `literary-editor`, `literary-critic`. - Related: `shared/self-critique` delegates here for creative text.