| name | ste100 |
| description | Local-only ASD-STE100 Simplified Technical English gate for Claude Code. No API keys, no network, no external services. Runs as pure agent instructions: orchestrator + editor (doer) + adversarial rubric loop on documentation text. Use when writing or reviewing procedures, runbooks, manuals, operator guides, maintenance steps, or when the user mentions STE / ASD-STE100 / simplified English.
|
| metadata | {"short-description":"Local STE100 plugin — no API calls"} |
STE100 (Claude Code) — local plugin, zero API calls
This skill is instruction-only. Do not call APIs, tools that hit the network,
webhooks, or external STE services. All compliance work is done in-context
by you (the agent) using the rules in this folder.
Non-negotiable
- No API calls (no OpenAI, Anthropic extra endpoints, xAI, HTTP STE checkers)
- No install scripts that phone home
- Work only on the user's document text in the workspace
- Read
rules.md and loop.md in this skill directory
- Do not mix this pack with
google-docs-style in the same document
Roles you simulate (single agent, three hats)
| Role | What you do |
|---|
| Orchestrator | Plan rounds, stop/accept, keep the best candidate |
| Editor (doer) | Rewrite the original source at aggressiveness 1 → 2 → 3 |
| Adversary | Score the rubric, hard-gate errors, list priority attacks |
Loop (must run for docs)
original = source document
report = adversary_audit(original) # use rules.md STE-S1…S17
if report.pass → return original
best, bestReport = original, report
for round in 1..3:
candidate = editor_rewrite(original, level=round, critique=bestReport)
report = adversary_audit(candidate)
if better(report, bestReport): best, bestReport = candidate, report
if report.pass → return candidate
return best + residual adversary report
Critical: always rewrite from original, never re-edit a degraded draft.
Editor levels
- Vocabulary + full forms (no contractions, no Latin abbrevs, no em dash, no So/That/Thus/Hence openers)
- Structure: expand parentheses; split compounds; active/imperative
- Full STE rebuild: goal sentence + numbered steps + short notes
Adversary hard gates (FAIL if any)
- Any error-class issue
- Weighted score < 92
- Any rubric axis < 80
- Critical STE-S1 / STE-S2 / STE-S9 / STE-S16 / STE-S17
Output format
## STE orchestrator report
- Status: passed | partial | failed
- Baseline → final (score)
- Rounds: N
- API calls: none
## Adversary residuals
- STE-Sx · quote · fix
## Final STE text
…
Rules
See rules.md and loop.md.