dev-team: trim filler from skill descriptions (−512 startup tokens)

[outofrange-consulting]

What

Conservatively trims 48 of the 79 dev-team skill descriptions. OMP renders each skill's name + description into the system prompt on every request (the /context "Skills" block, ~7.3K with dev-team). This cuts boilerplate ("Use this skill whenever the user wants to…"), synonym pile-ups, and redundant examples — keeping every trigger phrase, so auto-triggering is unchanged.

Measured (OMP 16.0.8 countTokens, live /context confirmed)

Skills block:   7,277 -> 6,765 tokens  (-512, -7%)
Startup total:   ~20K  ->  ~19.5K       (on top of the lean tool surface in #3)

• All 86 SKILL.md frontmatter blocks still parse as valid YAML.
• Only description changed — name / role / user-invocable / argument-hint / allowed-tools and skill bodies are untouched.
• Trigger-phrase preservation was verified programmatically (every quoted phrase in the original is retained).

Example — docker-image-audit (746 → ~430 chars): the six-way non-quoted synonym pile-up ("check a Docker image for security issues, scan a container for vulnerabilities, audit a Dockerfile, harden…, reduce image size, minimize attack surface, check for CVEs") collapses into the quoted trigger list that was already there. No behavior change.

Honest note on size

This is smaller than the ~1.7K I first projected. Conservatively, most of a description's length is trigger phrases (kept) plus substantive detail; the real filler — and so ~330 of the 512 tokens — lives in just the ~8 most verbose skills (competitive-analysis, ubiquitous-language, docker-image-{audit,create}, semantic-duplication-scan, js-project-init, semantic-scan, mermaid-diagramming). Pushing past ~512 would mean thinning trigger-phrase lists, which trades away triggering reliability — deliberately not done here.

Side benefit: the descriptions are now consistent and free of the repetitive "Use this skill whenever the user wants to…" lead-in.

🤖 Generated with Claude Code

https://claude.ai/code/session_01D98EuF8WovpMNRMbXBeg52

---

Generated by Claude Code

[outofrange-consulting]

Closing this as obsolete, on two counts.

1. The premise no longer holds on current OMP. This PR's saving rests on "OMP renders each skill's name+description into the system prompt on every request (the /context Skills block, ~7.3K)", measured on OMP 16.0.8. On OMP 16.1.16 (verified via a real install + interception of the actual launch request), the assembled context contains no skill names/descriptions at all — skills are now discovered on demand (pull-based via search_tool_bm25), not front-loaded. Measured launch context is ~14k tokens, 100% OMP core system prompt (~4.3k) + built-in tool schemas (~9.7k); the dev-team plugin contributes ~0 at launch. So the "−512 startup tokens" this PR targets no longer exist on current OMP.

2. It conflicts with merged work. The branch is based on 14d6da43 (pre-dating #7→#16) and edits descriptions of skills that #16 has since deleted (e.g. competitive-analysis) as part of removing plugin-meta skills (79→64). It would not merge cleanly.

The only residual benefit — less verbose descriptions for cleaner on-demand skill selection — is marginal and already overlaps the #16 cleanup. If we want it, it's cleaner to redo a small description-tidy fresh against current main than to rebase this. Closing.

---

Generated by Claude Code