dotfiles

Navigate and map large unfamiliar codebases. Find where behavior lives, trace feature flow and call chains, map architecture and dependency graphs, plan migrations/refactors, or answer "where is X" and "how does this work" questions. Also use for codebase scouting, hunting for specific patterns, discovering module relationships, or code archaeology. Activate on "find the code for", "how is X implemented", "navigate the codebase", "trace this", "map the architecture", or any task in an unfamiliar repo.

Human-inspired working + consolidated memory for agents, and the same progressive-disclosure pattern applied to docs and code. Two-tier store: unbounded short-term notes (working memory, per-branch session state, scratchpads) and size-limited long-term memory (curated facts, decisions, constraints, corrections) reached through an INDEX. A dream cycle runs on explicit request or git commit to consolidate hot short-term entries into long-term, score existing long-term entries, and propose evictions for human approval. Session checkpoints are one kind of short-term memory entry. Trigger phrases: 'remember this', 'save memory', 'checkpoint', 'save handoff', 'save context', 'what was I working on', 'resume', 'restore', 'load context', 'consolidate memory', 'dream cycle', 'prune memory', 'forget', 'evict', 'index my notes', 'structure docs', 'shard doc', 'split module', 'progressive disclosure'.

Coordinate a multi-persona deliberation session to challenge assumptions, debate architectural decisions, and perform pre-mortems. Leverages concurrent delegated agents/workers when the environment supports them, or falls back to single-context simulated dialogue. Trigger phrases: 'party mode', 'deliberation', 'multi-perspective', 'get a second opinion', 'debate this', 'challenge design'.

Bootstrap, audit, evolve, and materialize the AI-augmented project foundation (AGENTS.md, GLOSSARY.md, .agents/rules+skills core pack, Makefile, architecture, quality gates). Load for new project setup, missing skills after bootstrap, foundation drift, sync .agents from global, evolve AGENTS/GLOSSARY/architecture to match the repo, or audit project foundation. Skip only when editing a single existing rule/skill body with no pack-level change.

Multi-lens review orchestrator for code, specs, architecture, config, docs, and infrastructure. Hybrid execution: direct technical review using sub-reviewer references, or delegate to independent agents/workers for multi-perspective, audit, or specialized reviews. Sub-lenses: code-quality, design-rigor, adversarial, security, edge-case, editorial. Activate on "review", "check", "audit", "challenge", "QA", or any explicit review request.

Construct a precise, structured delegation prompt before every delegated-agent or worker launch. Use whenever you are about to delegate work to a background agent, run a task in parallel, or spawn a worker. Injects scope constraints, a structured output contract, obstacle reporting, and allowed-action boundaries directly into the input prompt to compensate for generic or unconfigured delegated-worker descriptions. Activate on: 'delegate this', 'run this in parallel', 'use the research agent', 'background task', 'spawn a worker', 'launch a subagent', 'hand this off'.

Swarminator orchestration: Mode Single-Node (one bounded external node, immutable artifact) or Mode Full Swarm (multi-phase multi-persona multi-model). Use for isolated research/second opinions/patch suggestions, or for diverse-perspective cross-validation when risk is too high for one pass. Activate on "swarm", "multi-agent", "external delegated node", "single-node", "swarminator", "second opinion", "cross-validate", "parallel agents". Commands: /swarm → Full Swarm, /external-subagent → Single-Node. Skip for routine single-agent implementation.

Generate or update docs/architecture.md with responsibility split, data ownership, data flow diagrams, API contract strategy, integration modes, and explicit non-goals. Load for "write architecture doc", "document system design", "create architecture diagram", "map data flow", or "define API contracts". Skip for minor code changes that don't affect architecture.

Bounded iterative execution for well-defined tasks with machine-verifiable completion criteria. Runs implementation → verify → retry cycles until success or a stop condition. Use for "run until done", "keep trying until tests pass", "iterate until it works", automated fix-verify loops, or any repetitive task with a clear pass/fail gate. Prefers human-in-the-loop by default; AFK mode requires isolation and explicit guardrails.

Facilitate creative ideation, product design, problem framing, and conceptual exploration. Use when generating new ideas, exploring solutions to open-ended problems, breaking creative blocks, or evaluating options from multiple perspectives. Trigger phrases: 'brainstorm', 'ideate', 'generate ideas', 'creative thinking', 'SCAMPER', 'six thinking hats', 'worst possible idea'.

Direct browser automation via Chrome DevTools Protocol — scraping, clicking, form-filling, login flows, file download/upload, DOM interaction, and end-to-end page testing without external services. Also covers web performance diagnostics: Core Web Vitals, Lighthouse audits, memory leaks, slow loads, long tasks, CDP traces, throttling, and network emulation. Activate on "open browser", "automate this website", "scrape", "click on", "fill this form", "download from", "test this page", "check performance", or any task requiring real browser interaction. Also activate when building site-specific browser knowledge (selectors, URL patterns, API endpoints).

Code design discipline enforcing SOLID, KISS, DRY, modularity, separation of concerns, and human-readability during implementation. Load for any non-trivial code write, feature addition, refactor, or restructuring. Covers: clean code, design patterns, decomposition, dependency management, naming, and architecture at the function/class/module level. Activate on "refactor", "restructure", "design this module", "clean up this code", "make this maintainable", "decompose", "extract", or any implementation touching more than one file.

Transform raw data into decision-ready narratives aligned to goals, KPIs, or OKRs. Multi-lens analysis with explicit evidence grading and confidence levels. Use for data analysis, metric summaries, trend interpretation, insight extraction, report writing, or presenting findings to stakeholders. Activate on "analyze this data", "summarize the metrics", "what do these numbers mean", "write a report", "present findings", "interpret trends", or any task turning data into actionable narrative.

Design CI/CD pipelines with integrated security scanning — change detection, environment separation, artifact retention, diagnostic collection, secret detection, dependency auditing, SAST, IaC scanning, and audit trail verification. Load for "set up CI/CD", "create pipeline", "configure GitHub Actions / GitLab CI", "automate deployments", "set up security scanning", "audit dependencies", "check for secrets", "security audit", "vulnerability scan", "compliance check", or "design deployment workflow". Skip for one-off script runs, local-only changes, or security reviews handled by the reviewer skill.

Full lifecycle skill authoring and maintenance: design and write new skills following the documented template, register them, run quarterly audits, review failure patterns, and guide skill evolution (prototype → harden → adopt → maintain → deprecate). Load for "create a skill", "write a new skill", "design a skill", "add a skill", "prototype a skill", "audit rules", "review skills", "maintain .agents/", "quarterly audit", "failure review", "deprecate skill", or "evolve rules". Skip for editing existing skills when no structural changes are needed.

Debug, investigate, and solve complex problems using structured frameworks (5 Whys, Fishbone, OODA, First Principles, PDCA). Use for tricky bugs, failure investigation, root cause analysis, architectural decisions, pre-mortem, or untangling multi-concern situations. Also activate when stuck in circles, on vague/recurring symptoms, or "why does this keep happening." Trigger phrases: "debug this", "troubleshoot", "figure out why", "root cause", "what's going wrong", "fix this bug", "this keeps breaking", "first principles", "pre-mortem", "red team this", "challenge assumptions", "go deeper", "rethink this", "inversion".

Screen CVs, evaluate candidates, and run professional OSINT verification/risk audits based on a job description and hiring context. Use when analyzing resumes, verifying candidate histories, researching hiring companies, or mapping market roles. Trigger phrases: 'screen CVs', 'screen candidates', 'evaluate candidates', 'resume screening', 'OSINT verification', 'candidate background check', 'talent screening'.

UI/UX design skill for building polished, accessible, production-grade interfaces. Covers layout, typography, color systems, motion design, responsive design, dark mode, glassmorphism, component styling, and visual polish. Uses taste sliders and audit-first methodology to prevent generic output. Activate on "design this UI", "make it look good", "improve the design", "frontend polish", "responsive layout", "add animations", "design system", "visual overhaul", or any UI/frontend design task.

AI video production pipeline for creating animated videos, motion graphics, and cinematic content. Structured workflow: style guide → story/script → asset extraction → shot list → render. Ensures visual consistency and cinematic quality across frames. Activate on "create a video", "animate this", "make a video", "storyboard", "motion graphics", "AI video", "film this concept", or any video/animation production task.

PRD, TRD, BDD specs, acceptance criteria, user stories, traceability, and changelog workflows. Use when the user requests requirements, product specs, technical design documents, architecture planning, behavior scenarios, scope definition, or a large ambiguous feature plan. Activate on "write a PRD", "define requirements", "acceptance criteria", "user stories", "feature spec", or any task needing formal specification before implementation.

Analyze and optimize agent model configurations based on live benchmark data, pricing catalogs, provider options, ELO ratings, and performance limits. Use when optimizing agent setups, conducting cost-benefit comparisons of LLMs, selecting model tiers for swarms, or validating model configurations against schemas. Trigger phrases: 'benchmark', 'optimize agents', 'model comparison', 'provider comparison', 'pricing', 'check model rankings', 'model benchmarking'.