Skip to content
Conduit

skill

dirigent

A Claude Code skill that routes subagent work across the opus/sonnet/haiku model matrix with a lead-session review gate.

xaversebastian/dirigentcommunitylistedunreviewed
Use skill

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/xaversebastian/dirigent itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 22, 2026

dirigent

A portable orchestration skill that turns the main session into a conductor. Non-trivial work follows:

Research → Plan → Act → Review

The lead researches the real state, decomposes the goal into evidence-bearing task blocks, routes eligible work by capability, and reviews every result before it counts as done.

What it gives you

  • A portable capability matrix (lead / reasoning-high / balanced / mechanical) in spec/capability-tiers.md.
  • Separate runtime adapters for each execution surface. Concrete mappings and their local verification basis live only in adapters/ and spec/mapping-table.md.
  • A dispatch discipline: supported overrides only, complete worker context, and parallel execution only for disjoint write scopes or read-only blocks.
  • A sequential fallback: if no compatible worker or override is available, the lead executes the same plan without fabricating worker results.
  • An escalation rule instead of blind retries (weak result → re-dispatch one tier up).
  • An operational review gate: every block has acceptance criteria and a done criterion; the lead checks tests, logs, paths, scope, side effects, and risks.

See SKILL.md for the full doctrine.

Install

Install the repository in the skill directory documented by the target runtime. For Claude Code:

git clone https://github.com/xaversebastian/dirigent.git ~/.claude/skills/dirigent

Then invoke it by name when you start a non-trivial task, or auto-inject it at session start via a SessionStart hook in ~/.claude/settings.json (see the "Installing" section in SKILL.md). Other runtimes should use their own skill location and invocation mechanism; an adapter does not imply hook support.

Notes

  • Capability tiers and workflow rules are runtime-neutral. Concrete model names are isolated to adapters and the explicit mapping document.
  • Adapter availability is checked against the versioned local-evidence fixture in tests/fixtures/runtime-models.json.
  • The distributed surface test works both in this OSS repo and in payload-only mirrors that intentionally omit maintenance files.
  • This is a routing layer, not an execution framework. It composes with whatever test/review/planning skills you already use rather than replacing them.

License

MIT — see LICENSE.

↑↓ navigate openesc close