trousse

MANDATORY gate BEFORE running jq on any .jsonl under ~/.claude/ or reading past CC sessions. Invoke FIRST when introspecting conversations, searching session history, parsing transcripts, or building tools that read ~/.claude/projects/ data. Provides the CC JSONL schema reference and `deglacer` CLI tool — prevents the 54-attempt fumble pattern where Claudes guess at field names. Triggers on 'what happened last session', 'find when we discussed', 'parse session', 'read conversation', 'session history', 'token usage', 'deglacer'. Do NOT use for git history (use git log) or your own current-session context. (user)

Inspects which ad, martech, analytics, and affiliate tags fire on any website and identifies each by vendor — tag names collide, so it confirms identity, not just labels. Drives a clean headless Chrome that bypasses ControlD DNS filtering and uBlock via DoH, captures the full network with cookies accepted, then groups every host by vendor and flags the unknowns for hand identification. Use when asked 'what tags fire on this site', 'is there a Meta, GA4, or Lantern tag here', 'audit this site for tracking', 'reverse engineer the martech', 'accept all cookies and show what loads', 'which vendors does this site use'. Triggers on 'tag explorer', 'tag scan', 'tamis'. For page reading or browser automation use passe; tamis is the tag-identification layer on top. (user)

Orchestrates cross-machine repo hygiene + GitHub account cleanup via an audit→approve→execute process that prevents accidental deletion. FIRST sweeps every git repo across machines (local + ssh) for uncommitted/unpushed work and true ahead/behind drift, THEN audits GitHub-side — Dependabot alerts traced to unused-direct vs transitive deps, stale forks, orphaned secrets, failing workflows, plugin version-bump gaps. Triggers on 'clean up GitHub', 'audit my repos', 'uncommitted or unpushed changes', 'are my repos in sync', 'push discipline', 'Dependabot trouble', 'unused deps', 'stale forks', 'dependency audit'. Requires gh CLI (+ ssh for remote hosts). (user)

Deep clean and structural health check for Claude-maintained codebases. INVOKE BEFORE adding significant complexity, WHEN inheriting an unfamiliar repo, or when something feels off and you can't name it. Three agents — Cracks (architecture), Dustballs (convention drift), Goofs (correctness) — examine in parallel, then synthesise into an honest assessment. Triggers on 'toise', 'deep clean', 'health check', 'should I be worried', 'check this codebase', 'is the architecture sound'. (user)

Spawns an isolated Claude with no CLAUDE.md, no skills, no hooks, no plugin context — only training weights and built-in skills. INVOKE BEFORE running evals, design research, cold critique, baseline comparisons, or testing plugin installs as a fresh user would experience them. Triggers on 'blank slate claude', 'fresh claude', 'isolated claude', 'naive claude', 'test plugin install', 'out of box claude', 'ardoise'. (user)

Searches Google developer documentation via REST API — returns fresh Markdown from sites like cloud.google.com, firebase.google.com, developer.android.com, ai.google.dev. BEFORE scraping or guessing at Google API docs, invoke FIRST. Triggers on 'check the Google docs', 'look up Firebase docs', 'what do the Cloud docs say', 'Google API documentation', 'search Google developer docs'. (user)

Transforms SPSS files, CSVs, and spreadsheet dumps into self-documenting BigQuery tables. Load BEFORE ingesting raw data — invoke first when you see opaque column names (S1, Q1r3), numeric codes, or empty BQ descriptions. Triggers on 'make this data analysis-ready', 'build a clean BQ table', 'load this SPSS file', 'enrich this table', 'process this codebook'. (user)

Spawn a peer Claude on the Anthropic conductor mesh to review code. The reviewer reads the target files, sends observations via mesh message, and you receive them as <channel> tags — peer-to-peer, not authority channel. Requires conductor-channel MCP server (aboyeur). Triggers on /peer-review, 'get a peer review', 'fresh eyes on this', 'spawn a reviewer'. (user)

Orchestrates new repo creation — required before setting up any batterie Python package. 5-step workflow (interview, plan, init script, CLAUDE.md fill-in, conformance validation) ensures every repo starts with hatchling, src layout, bon, and working tests. Handles greenfield, adopt, and extract modes. Triggers on 'create a repo', 'new project', 'scaffold', 'init repo', 'set up a new package', 'make this a proper repo'. (user)

Orchestrates all skill development — required before writing or editing any SKILL.md file. Unified 6-step workflow with automated validation, CSO scoring, register checks, and subagent testing. Triggers on 'create skill', 'new skill', 'validate skill', 'check skill quality', 'improve skill discovery', 'check this skill', 'write SKILL.md', 'edit SKILL.md', 'update skill description', 'can I share this', 'scan for sharing'. (user)

Orchestrates diagram creation with iterative render-and-check workflow. Invoke FIRST when asked to 'create a diagram', 'Venn diagram', 'flow chart', 'architecture diagram', 'visualize this' — ensures CRAP-principled self-critique before showing user, preventing amateur layouts and low-contrast text. Composes with brand skills for styling. (user)

Generate images using Google Imagen. Use when user needs 'hero image', 'presentation visual', 'conceptual illustration', or 'generate an image'. Composes with itv-styling for brand-constrained outputs. (user)

Take screenshots to see what's on screen. Triggers on 'screenshot', 'grab a screenshot', 'have a look', 'can you see', 'what does it look like', 'check the screen', 'did that work', 'verify it worked', 'what happened'. AFTER uncertain CLI operations (backgrounded processes, nohup, visual changes), consider capturing to verify state. Captures windows or full screen to files. (user)

Systematic Linux server management. Use BEFORE making changes to capture baseline, then AFTER for verification. Triggers on 'check this server', 'audit', 'set up this machine', 'security audit', 'harden this Pi', 'fresh Pi setup', 'provision this server'. (user)

Three-lens code review using parallel subagents: Epimetheus (hindsight — bugs, debt, fragility), Metis (craft — clarity, idiom, fit-for-purpose), Prometheus (foresight — vision, extensibility, future-Claude). Triggers on /titans, /review, 'review this code', 'what did I miss', 'before I ship this'. Use after completing substantial work, before /close. (user)

Writes and validates iA Presenter markdown where tab-indentation controls slide visibility. Triggers on 'presentation', 'iA Presenter', 'slides from markdown', '.presenter file', or when working in iCloud~net~ia~presenter folder. Validates tab characters (not spaces) for visible text,