Production-grade mods and a fully validated skills library for Command Code — battle-tested in daily use on a large monorepo, validity-tested before publication.
| Path | Contents |
|---|---|
mods/ |
12 TypeScript mods: workflow enforcement, turn pipeline, guards, proxies, telemetry |
skills/global/ |
37 skills, Anthropic Agent Skills spec, machine-audited |
docs/MODS.md |
Per-mod reference — hooks, config, ops notes |
docs/SKILLS.md |
Graded skills catalog + chain graph |
docs/ARCHITECTURE.md |
How it all composes: gate workflow, turn pipeline, proxy stack |
- Dream Gate — structured editing workflow: correlate → one batch plan (≤5 files) →
batch edits. Ships its own
correlatetool: lazy per-language import index (py/ts/js at boot, 6 more languages lazy), resolution-verified edges, depth-1–3 blast-radius trees. Validated to 100% recall / perfect precision against independently computed ground truth. - Resilience layer — crash-guard absorbs transport failures mid-session; keepalive retries router connections; loop-guard breaks deterministic tool loops.
- Skills library — every skill passes: parseable frontmatter,
name==dir, realchains_withedges only, referenced files exist, 80–200 line bodies with depth pushed toreferences/.
# Mods: copy what you need, then verify registration
cp mods/*.ts ~/.commandcode/mods/
cmd mods list
# Skills: global scope
cp -r skills/global/* ~/.commandcode/skills/Some mods expect local companions (proxy map file, Pieces LTM server) — see docs/MODS.md for per-mod requirements.
- Mechanical skills audit: 37/37 pass (frontmatter, naming, chains, references, line band)
- Content-density review across all skills (protocols, runnable code, worked examples)
- Correlate validity campaign: ground-truth fixtures, adversarial collisions/cycles, real-repo recall/precision audit vs independent truth set
- Sanitization scan of this tree: personal identifiers, private-theory branding, secret-shaped strings, compiled caches — all zero
MIT