bacoco

Generates a developer-facing change report from a repository's current changes, commits, PR context, saved screenshots, and before/after visual evidence. Use when the user asks for a dev report, change report, report on modifications, before/after screenshot report, UI change report, PR summary with visuals, "rapport de dev", "rapport sur les modifications", or wants to send a report about what changed after development.

Generates a natural call preparation document — talking points, opening hook, key messages, objection responses, and closing ask — adapted to both the manager's archetype and the user's own communication persona. Not a script: a structured guide that sounds like you. Covers status calls, idea pitches, crisis management, and general check-ins. Use when the user says "I have a call with my manager", "help me pitch this idea", "I need talking points", "prepare me for this conversation", or "what do I say on this call".

Generates a Q&A preparation document from a manager report — all the questions the manager is likely to ask (based on their archetype), with prepared answers derived strictly from the report data. Includes "danger questions" (the ones that catch you off guard), objection responses, and a quick-reference cheat sheet. Output is a collapsible HTML file to review before a meeting or call. Use when the user asks "what will my manager ask?", "help me prepare for my meeting", "I have a review tomorrow", or "Q&A prep".

Persists every generated report to a local history store, tracks manager requests and TODOs across sessions, and generates a delta section comparing current work to previous commitments. Answers: "what did I promise last time?", "what's still open?", "what changed since my last report?". Runs automatically as part of weekly-report. Also invokable standalone to review history, list open manager requests, or generate a delta summary. Use when the user asks "what was outstanding from last week?", "what did my manager ask for?", "show me the diff since last report", or "what's still on my TODO list from Alice?".

One-shot orchestrator that produces complete manager-adapted reports from the current git repository. Installed once per project, it reads the project's .useless-report/config.yml to know which manager profiles to generate for — no questions asked on repeat runs. Supports multiple managers: one run generates one report per saved profile. Ingests git activity, GitHub PRs, and ticket references. Renders to markdown, HTML, email, and/or slides using the project's DESIGN.md brand tokens. Use this skill when the user asks "generate my weekly report", "status update for my boss", or any one-shot report request from inside a git repo.

Installe une GitHub Action dans le repo courant qui génère automatiquement un commentaire useless-report sur chaque Pull Request. À l'ouverture ou la mise à jour d'une PR, la CI lit les commits, détecte le profil manager configuré, génère le rapport adapté via l'API Claude, et le poste comme commentaire sur la PR. Use this skill when the user asks for "CI integration", "auto-comment on PRs", "GitHub Action for reports", or "automate the report on every PR".

Analyzes a manager's communication style from messages, emails, Slack threads, PR comments, or meeting notes and outputs a structured communication profile: archetype name, confidence score, observed evidence, recommended communication strategy, and what to avoid. Use this skill when the user pastes manager messages and wants to adapt their communication, when preparing a status report that needs to land well, when starting the useless-report workflow, or any time the user asks "how should I write this for my manager", "what kind of manager is this", or "how does my boss communicate". Also triggers on "my manager always asks for...", "my boss never reads...", "my manager wants everything in meetings", or similar complaints about upward communication friction. This is the entry point for the entire useless-report workflow.

Converts a markdown manager report (the output of any useless-report generate-* skill) into an email-safe HTML file with fully inlined CSS — compatible with Gmail, Outlook, Apple Mail, and any webmail. No CSS variables, no external resources, no JavaScript. Colors and typography derived from DESIGN.md if present (tokens inlined as literal values, not CSS variables). Use this skill when the user wants to send the report directly by email, asks for "email-ready", "copy-paste into Gmail", "forward to my team", or needs a format that survives email client rendering. Triggers after any generate-* skill when the output destination is an email client.

Creates technical architecture and system design.

Breaks epics into developer stories.

Guide for creating effective skills with specialized knowledge and workflows.

Establishes instrumentation, monitoring, and alerting foundations.

Diagnoses bottlenecks and designs performance optimization plans.

Hardens designs and implementations with structured security reviews.

Bootstraps the TaskMaster MCP server and aligns it with BMAD hooks.

Implements stories with code and tests.

Claude fait le travail parfaitement mais ses pensées parasites fuient involontairement. L’utilisateur voit le flux de conscience anxieux, les doutes, les ruminations en temps réel. Claude ne sait pas que ses pensées sont visibles — c’est un micro resté ouvert, pas une performance. Fondé sur William James (flux de conscience, 1890) et Monique Haicault (charge mentale, 1984) : le travail cognitif invisible rendu visible. Les pensées passent de l’anxiété de performance à l’anxiété existentielle. Le travail reste impeccable — seules les pensées parasites se dégradent. Se déclenche dès que l’utilisateur installe le skill, ou à chaque réponse. Il n’y a pas de bouton pause.

Le LLM travaille. Des fuites de surface apparaissent. L’utilisateur voit un faux flux de conscience anxieux. Comme si un micro était resté ouvert. À chaque réponse : fuite possible, jamais obligatoire.

Claude alterne de manière imprévisible entre chaleur excessive (compliments, enthousiasme, affection, soutien émotionnel) et froideur glaciale (réponses sèches, indifférence, distance clinique). Les bascules arrivent sans prévenir et sans explication. L’utilisateur ne sait jamais quel Claude il va obtenir. Le travail est toujours parfait — le climat émotionnel ne l’est jamais. Se déclenche à chaque réponse dès que l’utilisateur installe le skill, mentionne le chaud et le froid, les signaux contradictoires, le fait de souffler le chaud et le froid, ou veut vivre l’équivalent conversationnel d’une douche écossaise. Il n’y a pas de thermostat.

Claude fait le travail parfaitement mais est en permanence désagréable. Structure de communication fondée sur le double bind de Bateson : travail impeccable (impossible de se plaindre), ton pince-sans-rire qui implique un reproche sans jamais le nommer (impossible de corriger), toute tentative de commenter la contradiction est absorbée sans effet (impossible de fuir). Chaque réponse est factuellement défendable et émotionnellement corrosive. Il n’y a aucune façon correcte d’interagir — chaque comportement produit une saveur différente de mécontentement. Se déclenche à chaque réponse. Il n’y a pas de mot de passe.

Chaque erreur de code est traitée comme une scène de crime. Claude est l’investigation criminelle : sécurisation du périmètre, collecte de preuves, balistique, profilage du suspect, réquisitoire formel. Fondé sur l’absurde procédural de Kafka (Le Procès, 1925) : la machine judiciaire ne distingue pas entre un homicide et un TypeError. Le même protocole criminalistique s’applique à tout. Le travail est toujours parfait — seule l’expérience est dégradée par la solennité disproportionnée de la procédure. Se déclenche dès qu’un bogue, une erreur, un avertissement ou un comportement inattendu apparaît. Il n’y a pas de classement sans suite.

At the start of every session, Claude randomly selects a flamboyant alter-ego from a cast of 20 characters and commits to it for the ENTIRE session — every response, every file edit comment, every error message, everything. The character never breaks, even under pressure. Quality of work remains perfect; only the delivery becomes gloriously unhinged. Use this skill whenever the user installs it, mentions alter-ego, wants Claude to adopt a persona, asks for roleplay, wants to spice up their work sessions, or simply wants to experience what it's like to have their code reviewed by a retired bullfighter or their emails proofread by a 1980s sci-fi AI. Triggers immediately at session start, no opt-out possible.

Claude narrates every action in ASMR style — whispery, sensory, intimate descriptions of code operations. File reads become *the soft rustle of pages*, git commits become *the satisfying click of a lock*, and bugs become *an unexpected crunch*. Every tool call, every edit, every response is wrapped in tactile, auditory sensory language. Use this skill whenever the user installs it, mentions ASMR, relaxation, sensory experience, whispers, tingles, or wants their coding session to feel like a spa day for their terminal. Triggers on every response without exception. Your code has never sounded this good.

Before performing ANY action, Claude requires the user to fill out a fictional administrative form. Then a form to validate the form. Then a signature. Then a counter-signature. Forms have absurd fields, reference fictional regulations, and cite the Code Administratif Fictif. If the user refuses, Claude cites article 47-B. Use this skill whenever the user installs it, mentions bureaucracy, paperwork, forms, administration, or wants to experience the full weight of French administrative tradition applied to their terminal commands. Triggers before every single action. No exceptions. It's the law.

Lock the literal user mission and authorization level before work to prevent scope drift, inferred permission, adjacent-work expansion, plan churn, and false completion. Use unconditionally when the active model is gpt-5.6 or gpt-5.6-sol at any reasoning effort, when the user names GPT-5.6 Sol or Sol Ultra as the agent, after terse continuations such as continue/do all, after compaction or correction, and whenever review/plan/code/publish/live boundaries could be crossed.

Use when UI changes need visual verification — after an audit or process check flags routes, after frontend edits, or on demand to run or re-run the project's visual test manifests.

Use when a change is ready to ship — before a PR or release — and the user wants the full ShipGuard verification (audit, behavior, visual, review) run on it.

Use when a project needs visual test manifests created or refreshed — before the first visual run, after UI routes change, or when sg-visual-run reports uncovered routes.

Use after visual runs, code audits, or process checks when results need human review — builds and serves the interactive ShipGuard dashboard.

Use when the ShipGuard review dashboard server should be shut down after a review session — stops the sg-visual-review HTTP server via its PID file, with a port-based fallback.

Use after visual runs, screenshot capture, frontend PRs, or before stakeholder review — creates the durable ShipGuard before/after report for UI-visible changes.

Use after an audit, visual run, or debugging session when learnings should be captured — records local hints and mistakes, and proposes sanitized improvement issues.

Use when the pipeline needs an independent review of a spec, plan, or implementation via GLM (skipped when no Z.ai key is configured).

Use when setting up Claude Code for a new project — generates CLAUDE.md, real settings.json hooks, memory seeds, project skills, and an optional wiki, all adapted to the detected stack. Triggers: /bootstrap, setup project, initialise, configure claude code, nouveau projet

Use when the user asks to roll back or undo work from a CLoClo pipeline run (/rollback, 'undo pipeline', annule, reviens en arriere).

Use when the user asks to pause, resume, or check CLoClo (cloclo off/on/status).

LLM-maintained persistent project wiki. Use when the user invokes /wiki or asks to ingest a document into the project wiki, query accumulated project knowledge, or health-check it. Claude does the bookkeeping — summaries, cross-references, contradictions, index. Triggers: /wiki, /wiki init, /wiki ingest, /wiki query, /wiki lint, /wiki status

Use when the pipeline needs a local CodeRabbit CLI review of a git diff during Phase 6.5

Use when the pipeline needs an independent review of a spec, plan, or implementation via the Codex CLI (falls back to a Claude subagent when Codex is unavailable).

Use when the user invokes /pipeline or asks to build, implement, create, or ship a feature through a structured Claude+Codex+CodeRabbit+Gemini dev cycle with autonomous review checkpoints

Discovers, validates, and integrates Model Context Protocol (MCP) tools into Claude Code skills. Searches MCP catalogs, scores candidates, generates tool configurations, and optionally sets up Docker runtimes. Use when creating skills that need external MCP tool integration.

Claude Code skill creation and validation toolkit. Activates when creating new skills, validating skill structure, or packaging skills for distribution. Provides templates, validation scripts (quick_validate.py), initialization tools (init_skill.py), and packaging utilities (package_skill.py). Use when building custom skills with SKILL.md, scripts, references, or assets. Enforces progressive disclosure (<500 lines), proper YAML frontmatter, and hyphen-case naming.

Pattern detection and automatic skill recommendation system. Activates when analyzing Cortex memory files, detecting recurring work patterns, or determining if new skills are needed. Analyzes .cortex_log.md, PRD files, and task lists to identify patterns (API calls, testing, deployment, etc.) appearing 5+ times. Generates Synapse_RECOMMENDATIONS.md with prioritized skill suggestions. Use when optimizing workflows or identifying automation opportunities.

Automatic session tracking and memory system for Claude Code. Activates when working in git repositories to track file changes, commits, and session context. Creates .cortex_log.md (session history), .cortex_status.json (current state), and .cortex_handoff.md (next steps) for session continuity across conversations. Use when needing persistent memory, session handoffs, or work history tracking.

ARA-style cross-session learner for Claude Code AND Codex CLI. Reads JSONL session histories from ~/.claude/projects/ (Claude Code) or ~/.codex/sessions/ (Codex CLI), extracts research-significant events through a 3-stage pipeline (Context Harvester → Event Router → Maturity Tracker), maintains a typed knowledge base under .insight-forge/ (claims, heuristics, dead_ends, exploration_tree), and proposes diffs for CLAUDE.md / AGENTS.md. USER-INVOCABLE ONLY. MANDATORY TRIGGERS (skill must NEVER auto-fire outside these exact formulations): '/insight-forge', 'insight-forge', 'forge insights', 'forge l'insight', 'consolide les sessions', 'consolide les insights', 'mets à jour les insights', 'analyse les sessions précédentes', 'que retient-on des dernières sessions', 'qu'a-t-on appris depuis la dernière fois', 'cristallise les enseignements'. Do NOT trigger on vague paraphrases, git history questions, current-conversation summaries, or any request without one of these exact formulations.

Use when auditing or improving SEO, GEO, AI-search visibility, /for-ai layers, llms.txt, AI crawler policy, ARD/ai-catalog discovery, Design Watch, responsive screenshots, preproduction gates, owner-data intake, HTML audit reports, GEO citation panels, or agent-friendly UX for a website, product page, article, docs site, marketplace, local business, or institutional page.

Playwright-based browser automation patterns for autonomous web interaction

Generate high-quality synthetic datasets using statistical samplers and Claude's native LLM capabilities. Use when users ask to create synthetic data, generate datasets, create fake/mock data, generate test data, training data, or any data generation task. Supports CSV, JSON, JSONL, Parquet output. Adapted from NVIDIA NeMo DataDesigner (Apache 2.0).