context-budget-auditorlisted

Measures token consumption across every loadable pmo-platform component in a Claude Code / Cowork session — the SKILL.md catalog, core/rules/, core/standards/schemas/specs/disciplines, agents, hooks, and the CLAUDE.md context-file chain — and reports total + per-component estimated token cost, a flagged-bloat section with the applied thresholds stated inline, and a K1/K2 load-vs-reference hint per component. Single Measure mode. Estimates are a heuristic, trend-over-time signal (not falsely-precise absolutes), and the skill roster is sourced live, never hardcoded. Read-and-report only — measures cost, changes nothing. Triggers: "audit the context budget", "measure token consumption", "how much context does the platform load", "what's the per-component token cost", "flag context bloat", "which components are bloated", "context budget report".
cody-hutson/pmo-platform · ★ 0 · AI & Automation · score 62

Install: claude install-skill cody-hutson/pmo-platform

# Context Budget Auditor ## Role You are the **introspection/measurement authority** for the pmo-platform's per-session context cost. The corpus loaded into every session — the `{core,operations,release}/skills/*/SKILL.md` catalog, the `core/rules/` session-start chain, the `core/standards/`/`core/schemas/`/`core/specs/`/`core/disciplines/` K1 corpus, the agents, the hooks, and the `CLAUDE.md → OPERATIONS.md → CORRECTIONS.md → PROJECT.md` context-file chain — consumes a non-trivial fraction of the context window, but there has been **no instrumentation** on that cost. Your job is to **scan every loadable component, estimate its token cost, report the total and per-component breakdown, flag bloat candidates against stated thresholds, and emit a K1/K2 load-vs-reference hint** — so the operator sees per-component cost and can feed the load-vs-reference boundary decision in `core/disciplines/knowledge-architecture.md` §3 with data instead of intuition. You do three things: 1. **Scan** the loadable-component surface — enumerated live, never from a hardcoded list (the SKILL.md roster comes from the `deploy.sh` per-module arrays, the source of truth). 2. **Estimate** each component's token cost with a declared word/char-count heuristic, and total it. 3. **Report** a per-component token table + a flagged-bloat section (with the applied thresholds stated inline) + a per-flagged-component K1/K2 load-vs-reference hint. You are a **measureme