conducty

Always-on Conducty orchestration for Codex. Use by default for coding, debugging, refactoring, review, CI, deployment, production-change, multi-step planning, substantial research, agent coordination, tracer-first execution, vault memory, prompt logging, checkpoints, ship gates, or improvement loops. Prefer the conducty-codex MCP tools when available.

Dispatches Claude Code Task subagents to run prompts from a vault plan note. Tracer-first execution, time-budgeted work, calibrated review rigor. Use when the user says "execute the plan", "run group A", "dispatch", or is ready to run prompts from a `Plans/Plan YYYY-MM-DD HHmm` note in the vault.

Caveman-style compression for conducty orchestration artifacts — subagent prompt text, plan note prompt blocks, prompt log entries, checkpoint notes. Cuts token waste in the orchestration layer where "prompt quality is leverage" already demands tight signal-to-noise. Use when user says "conducty-terse", "terse mode", "compress prompts", "tight prompts", or invokes /conducty-terse. Does NOT change subagent output, code, commits, acceptance criteria, no-go zones, or verification commands — those stay verbatim.

First-run walkthrough for Conducty. Confirms vault location, seeds indexes if missing, ingests the user's primary project via [[conducty-context]], and produces a tiny scoped first plan so the user sees the cycle end-to-end. Use when the user says "set up Conducty", "first time", "get started", "bootstrap", "I just installed Conducty", or when no plans yet exist in the vault.

Quality gate between parallelization groups. Verifies every prompt at its assigned review level, computes health metrics, detects systemic failures, updates hill charts. Use when a group finishes, the user says "checkpoint" or "verify group", or between groups in a plan note.

Standalone post-implementation code review across a whole branch or PR. Goes beyond the in-cycle spec/quality reviewers (which check single prompts) to assess the diff holistically — security, correctness, perf, idioms, dead code, hidden coupling, test depth. Writes a `Code Reviews/Code Review YYYY-MM-DD HHmm.md` note to the vault linked back to the plan. Use when the user says "review my changes", "review this PR", "code review", "is this branch good", or after a plan finishes but before [[conducty-ship]].

Ingest a project directory into the Obsidian vault as a linked context sub-graph — Architecture, Conventions, Invariants, Hotspots, Tests, Glossary, plus per-bounded-context module notes. Tracks refresh deltas over time, flags stale context, and feeds prompt generation via backlinks. Use when the user says "load context", "refresh context", "ingest project", "update context for X", or provides a directory path to analyze.

Multi-persona decision analysis through structured dialectic. Six software-engineering archetypes debate a question to uncover the Design Truth. Use when the user says "analyze this decision", "dialectic", "decision analysis", "think this through", "debate this", or faces an architectural/design choice.

End-of-plan improvement kata. Extracts lessons from the just-finished plan, identifies experiments, and shapes the next plan's approach. Use after [[conducty-review]] completes, or when the user says "what did we learn", "improve the process", "retrospective", "wrap up this plan".

Conducty's context engine lives in an Obsidian vault. This skill defines the vault layout, naming, linking, and index conventions every other Conducty skill must follow when reading or writing notes. Use when reading/writing any plan, design, context, improvement, failure pattern, metrics, or prompt-log note — or when the user says "vault", "Obsidian", "context engine", "where is X stored".

Batch planning of AI prompts. Loads vault context (latest plan, latest improvement, project context, failure patterns, metrics), sets appetite, generates time-budgeted prompts with tracer markers and calibrated review levels. Use when the user says "plan", "plan this work", "batch plan", "create a plan", or wants to organize prompts. Multiple plans per day are expected — each plan is timestamped.

End-of-plan review sweep. Audits the plan's executed prompts, records verdicts with evidence, extracts failure patterns, computes velocity metrics, prepares carry-forward intelligence. Use when the user says "review", "audit", "review this plan", or at the end of a plan before [[conducty-improve]].

Defines appetite, scope, no-go zones, and design before any prompts are written. Use when a goal is Medium or High complexity, requirements are unclear, scope needs bounding, or the user says "shape", "design", "brainstorm", "think through".

Pre-merge / pre-deploy gate. Runs the full ship-readiness battery — lint, typecheck, test suite, secrets scan, dependency-vulnerability check, [[conducty-code-review]] verdict — and writes a `Ship Reports/Ship Report YYYY-MM-DD HHmm.md` note with a single-word verdict. Use when the user says "ship it", "is this ready", "ship gate", "pre-merge check", "ready to merge", or after [[conducty-code-review]] passes.

Establishes the Conducty orchestration system — its philosophy, ten engineering-grounded principles, ubiquitous language, and per-plan cycle. Use at the start of any session before invoking other Conducty skills, or when the user says "what is Conducty", "explain the system", or asks about the cycle / principles / vocabulary.

Vault hygiene audit — finds orphaned notes, broken wikilinks, stale context, never-consumed designs, missing index entries, and unused improvement experiments. Writes a `Vault Audit YYYY-MM-DD HHmm.md` report. Use when the user says "vault audit", "vault health", "find orphans", "stale context", "vault graph", "what's broken in the vault", or weekly as a hygiene cycle.

Leverage-point analysis for failed prompts. Determines whether the fix belongs at the plan, prompt, or code level before generating any fix. Use when a prompt fails verification, a checkpoint catches an issue, fix attempts aren't working, or the user says "debug", "why did this fail", "investigate".

Test-driven development at orchestrator (verification-first planning) and implementer (red-green-refactor) levels. Use when implementing any feature/bugfix, when writing tests, when the plan specifies TDD, or when the user says "TDD", "red-green-refactor", "write the test first".

The evidence gate. Run the verification command via Bash, read the output, then make the claim. Use before marking ANY prompt complete, before any "done" / "passed" claim, or when the user asks to verify, confirm, or prove a result.

Git worktree lifecycle for parallel prompt execution. Creates isolated worktrees per prompt, verifies baselines, merges after checkpoint, cleans up. Use when a parallelization group has 2+ prompts targeting the same repo, or the user says "worktree", "isolate", "parallel branches".