| name | ste-writing |
| description | Rewrite hand-authored durable prose (ADRs, CONTEXT.md, CONTEXT-MAP.md, README.md, OpenSpec artifacts, PR descriptions, release notes) toward ASD-STE100 Simplified Technical English to remove AI slop. Does NOT apply to code, identifiers, command syntax, docs/ARCHITECT.md (code-derived), commit messages (owned by caveman-commit), or chat (owned by caveman). |
Two modes
STE-flavored (default). Overridable editorial heuristics. Applied to ADRs, CONTEXT, CONTEXT-MAP, README, OpenSpec artifacts, PR descriptions, release notes.
Strict. Mandatory. Apply strict mode only to procedure-class prose: operator runbooks, migration/remediation steps, safety/security instructions, actionable error-message guidance.
Self-lint rules
Six mechanical rules:
(a) Split sentences over 20 words.
(b) Replace semicolons with periods.
(c) Expand contractions.
(d) Make passive voice active when the actor is known.
(e) Replace -ing main verbs, nominalizations, and phrasal verbs with plain verbs. Examples: "perform an analysis" -> "analyze", "spin up" -> "start".
(f) one name per concept.
In flavored mode these are overridable. In strict mode they are mandatory for prose.
Slop markers
Flag these as AI slop:
- Banned verbs: "leverage", "utilize", "facilitate", "ensure", "prior to".
- Marketing adjectives: "seamless", "robust", "powerful".
- modal hedge: "it is important to note".
- em-dash (an LLM tell).
Code protection
Code spans, fenced blocks, identifiers, and verbatim commands are never rewritten in either mode. Lint rules do not fire on them.
Voice preservation
In flavored mode the skill preserves canonical terms, causal argument, deliberate emphasis, and project-defining formulations (e.g. "Every processor and producer is a Service<Exchange>"). STE clarifies prose. It does not strip voice.
Surface division
caveman owns chat. caveman-commit owns commit metadata. ste-writing owns durable explanatory, procedural, and operator-facing prose. No two run over the same text at once.
