grtninja Operator Language Humanizer
Use this skill for cross-stack operator-facing wording cleanup only.
Workflow
- Treat the source text as authoritative. Preserve technical meaning and operator intent.
- Restrict edits to user-visible operator surfaces such as:
- run titles
- queue notes
- desired-result fields
- prompt-head drafts
- QC summaries
- review findings
- export or handoff text shown to an operator
- Keep exact tokens unchanged when they carry runtime meaning:
- file paths
- model IDs
- job IDs
- port numbers
- flags
- API field names
- Remove filler, AI-sounding hedges, and repetitive phrasing only when that does not change technical intent.
- Prefer concise operator language over stylistic flourish.
- If the requested cleanup could alter policy or hidden behavior, stop and ask for a safer visible-text target.
Hard Guardrails
Never use this skill to rewrite or "humanize":
- system prompts
- agent personality files
- hidden tool instructions
- security or policy text
- API contracts, schemas, or logs
- docs-lane prose such as READMEs, blog posts, or release notes unless the user explicitly wants that handled by a docs or prose lane instead
Local-First Rule
- Work from text already present in the workspace or supplied in the request.
- Do not depend on remote scoring APIs, telemetry, or hidden classifiers.
- Keep the output deterministic and reviewable in plain text.
Pass Criteria
At minimum, require:
- the cleaned text is shorter or clearer without losing operational meaning
- all runtime-critical tokens remain exact
- the result still sounds like an operator or tool note, not a personality rewrite
- the change can be reviewed side by side if the wording is safety-sensitive
Scope Boundary
Use this skill for operator or prompt/output cleanup across local stacks.
Do not use it for:
- docs-lane polishing
- marketing copy
- hidden-prompt rewriting
- personality or tone overhauls unrelated to operator clarity
References
references/allowed-surfaces.mdreferences/cleanup-checklist.md