@camacho

local-merge safely lands branches onto targets by merging in a disposable shallow clone first, then updates the primary worktree only if already on target, with explicit handling for conflicts, divergent commits, and tracked state preservation.

test-driven-development enforces writing failing tests before implementation code, with explicit red-green-refactor cycles. covers the iron law of tdd, common rationalizations, and verification checklists for features and bug fixes.

isolate creates a fresh git worktree branched from origin/main, detecting the parent branch via reflog, auto-installing dependencies, and running baseline tests before task work begins.

imperatives extracts RFC 2119 keywords and implicit obligations from markdown rule files via dual-pass analysis (script + subagent reasoning), outputting structured jsonl for downstream policy tooling.

policy-algebra generates frozen starlark governance rules from intent strings or plan files, supports drift detection via --verify, and falls back from shallow (file-based) to deep (interview-driven) extraction when invariants are sparse or ambiguous.

build executes approved plans via parallel subagent dispatch, batching tasks by dependency, validating output, and enforcing commit boundaries between batches to prevent loss of work.

writing-plans structures multi-step task implementation for engineers unfamiliar with the codebase, breaking specs into bite-sized, testable tasks with file-level detail and testing guidance.

requesting-code-review dispatches a focused reviewer agent with git context between two commits, establishes when to request review (mandatory after tasks, before merge; optional when stuck), and prescribes triage of feedback by severity.

catchup reconstructs branch context after interruptions by gathering git state, workflow metadata, plans, and github status, then synthesizing findings into actionable next steps.

session-start-hook provides structured input schema and event wiring for Claude Code initialization, enabling dependency installation across CLI/web/API contexts. Execution logic and recovery paths remain incomplete.

detects new or modified skills in .agents/skills/ and publishes them to ai-skills after git validation, with snapshot capability for rollback and manual review before deployment.

to-issues decomposes plans and specs into vertically-sliced, independently-grabbable issues using gh and project-specific vocabulary. focuses on tracer-bullet delivery over horizontal layering.

bail detects workflow step by filesystem state, reflects on exit reason, updates github issue, closes open pr, and cleans up worktree and branch artifacts.

orient fetches github issue context and auto-detects task type from labels to map a git branch prefix, accelerating task initialization and branch naming consistency.

wrap coordinates session closure by invoking reflect to finalize memory state, then cleans git worktrees and prepares context for the next task cycle.

plan-review enforces a gate at the plan checkpoint, assembling a deterministic review panel and collecting approval verdicts before build phase entry. incomplete definition of the review logic, agent dispatch, and verdict collection leaves execution ambiguous.

assembles a panel of reviewer agents and enforces governance policy for code review loops. takes scope (file path or diff) and optional include/exclude overrides, returns ordered reviewer list plus gate/cap parameters.

position loads a teammate profile from disk and bootstraps session identity. it distinguishes between persistent positions (tpm, engineer) and transient roles, with position-specific state directory binding.

archive captures outcomes and learnings from plan files by prompting the user to reflect on what worked, what failed, and what to change next time, then renames the plan to .done.md.

review auto-detects plan or code context, gates by severity threshold, and runs bounded fix loops. mode detection heuristics handle common workflows but lack clarity on ambiguity recovery.

reflect consolidates session work into memory by reviewing git history, closing linked issues, and verifying phase gate compliance before declaring task completion.

copilot mode switches claude into human-driven collaboration by relaxing worktree constraints and enabling main branch commits, with activation via shell hook.

exits copilot mode by activating a project-local hook script, with fallback messaging for missing configurations. intended for claude.ai sessions where mode switching is relevant.

status collects git metadata and project telemetry into a read-only snapshot. incomplete skill definition lacks decision logic, output contract, and outcome signals.

validate executes a single pnpm command wrapping typecheck, lint, format, test, and optional build stages, with minimal guidance on failure modes or recovery.

task is a routing table that maps work types to tool recommendations, not a procedural dispatcher. lacks executable steps, failure handling, and concrete decision criteria.

workflow is a passthrough alias to /task with no independent logic, documentation, or actionable guidance. delegates all responsibility upstream.

Use when implementation is complete, tests pass, and code needs to reach main — via PR with auto-merge (CI up) or local shallow-clone merge (CI down)

Creates a draft GitHub Issue with triage label from natural language description.

Triage issues through a state machine driven by triage roles. Use when user wants to create an issue, triage issues, review incoming bugs or feature requests,…

Use when pulling/syncing/updating/absorbing changes from a branch's parent (typically main) into the current branch, or cascading parent changes through a…

brainstorming — an installable skill for AI agents, published by camacho/ai-skills.

to-prd — an installable skill for AI agents, published by camacho/ai-skills.

Use when creating or rewriting a skill and you want one strong, workflow-conformant version built through failure analysis, review, and refinement instead of a…