Skip to content

undefined-ui/second-brain-os / plugins/agents-course/skills/context-audit

context-audit

Unverified

Audit an agent's context layout against the four places: system prompt, tools, history, tail. Use when the user asks to audit, review or fix their agent's context, prompt caching, token spend per turn, or CLAUDE.md / AGENTS.md layout. Read-only: reports findings and a fix list, changes nothing. Do NOT use for writing evals (evals-bootstrap) or for loop design (goal-test).

Audit the context window

Theory: How models read context and The four places. Every piece of context belongs in exactly one of four places — a byte-stable system prompt, a frozen tool set, a compacting history, and a short live tail — and most agent problems trace back to something sitting in the wrong one.

Core rule

Report and rank; never edit. The output is an audit, and the user decides what to apply. If they ask you to apply fixes afterwards, that is a normal edit session, not this skill.

Workflow

  1. Find the context sources. Locate what actually reaches the model: system prompt (or CLAUDE.md / AGENTS.md for a Claude Code setup), tool definitions, and — if the project logs requests — one full mid-conversation payload. If nothing is logged, say so and audit the static files; ask for one captured request only if the user can produce it cheaply.
  2. Measure the five numbers. Approximate token counts for: system prompt, tool definitions, message history, retrieved content, current query. A rough count (chars / 4) is fine; write all five down.
  3. Hunt cache killers. Flag every dynamic value in the system prompt — timestamps, user names, injected memories, "current task" lines. Each one is a guaranteed cache miss on every call. Check whether the tool list can change mid-run; a changing list invalidates the whole prefix.
  4. Hunt window bloat. Find the largest single item in the history. A verbatim tool result over ~2,000 tokens should have been a file path plus a one-line receipt. Count tools: past twenty (or ~10K tokens of definitions), recommend deferred tools + tool search.
  5. Check the tail. Locate the current goal. If it appears only in the opening message, it lives in the weak middle of the window — recommend restating it in the tail every three to five steps.
  6. Check guides. If CLAUDE.md / AGENTS.md exists: it should be a few lines of constraints the code cannot show, not four pages of narrative. Flag anything the repo already records (structure, history).

Output format

Context audit — <project>
system prompt   <n> tok   <clean | N dynamic values: ...>
tools           <n> tok   <n> tools  <frozen | mutates mid-run>
history         <n> tok   largest item: <what, n tok>
tail            <present | goal only in opening message>

Top fixes, in order of saved tokens per turn:
1. ...
2. ...

Each fix names the file and line where possible, states what moves to which of the four places, and estimates the saving. Close with the one-line rule: stable prefix, frozen tools, compact history, live tail.

rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 192kUpdated Claude CodeCodexVerifiedInstall