claude-agent-dev

Performs architecture review and system design for problems spanning multiple files or crossing module boundaries in the target repo's own code. Analyzes existing directories, dependencies, and imports to generate an Architecture Decision Record (ADR) in docs/adr/ and an architecture-brief.json schema. Trigger on: 'architecture review', 'restructure across modules', 'too coupled', 'design this system', 'where should this code live', 'God class', 'circular deps', 'dependency mapping', 'domain boundaries'. Also triggers when addressing boundary violations, structural bleed, or resolving high git coupling between files. Always prefer this skill over refactor or planning when designing multi-module interfaces or correcting structural coupling before writing detailed implementation steps. Defers to make-a-skill for auditing or designing the skills/ directory itself.

Conducts collaborative discovery, requirements analysis, and system brainstorming before initiating creative implementation or feature building, producing a markdown-kv Design Brief in docs/design/ and session logs in .brainstorm/. Trigger on: 'design a feature', 'add a feature', 'brainstorm a solution', 'write a design doc', 'conduct discovery', 'build a new feature', 'requirements discovery'. Also triggers when request details are ambiguous, require terminology clarity in a glossary, or have unstated stakeholders. Always prefer brainstorming over planning or architecting when requirements are still vague, unproven, or not yet codified into a clear technical specification.

Initialize or audit repository instructions. Generates high-signal AGENTS.md wired to CLAUDE.md/GEMINI.md stubs. Not for documenting a specific feature's spec or design (see planning, brainstorming). Trigger on: 'init codebase', 'onboard repo', 'audit AGENTS.md', 'setup agent instructions', 'codebase-init', 'initialize project memory'.

Diagnoses conversation bloat and prunes active context using token diagnostics, rolling summaries, and script-based compaction. Generates a flat key-value state summary file before executing the `/clear` command to reduce active token usage. Trigger on: 'optimize context', 'compress context', 'prune memory', 'reduce tokens', 'context size too large', 'out of tokens', 'compact context', 'clear history', 'cleanup context'. Also triggers when managing long development sessions, handling large file modifications, or resolving memory drift in nested subagent workflows. Always prefer this optimizer over running `/clear` directly when critical task progress and decisions need to be preserved.

Performs disciplined root-cause analysis for software bugs and crashes using a systematic hypothesis-falsification workflow. Accepts a symptom description or error trace, and produces a structured diagnosis report detailing the symptom, root cause, fix, and reproduction feedback loop. Trigger on: 'debug', 'fix crash', 'why is this failing', 'unexpected output', 'diagnose bug', 'root cause analysis', 'feedback loop', 'instrumentation'. Also triggers when troubleshooting test failures or diagnosing unexpected runtime exceptions. Always prefer this skill over test-driven-development or refactor when diagnosing a bug prior to implementing a fix.

3 hooks across 3 events

Design → Build → Validate → Ship. Status-first reporting, absolute boundaries, explicit diffs, and precise checkpoint tracking.

Skill script tests — live feedback during skill development

Hook execution telemetry — surfaces errors and slow hooks in real time (cross-platform)