scriptoriumadmin/claude-skills
1
0
Fork 0
mirror of https://github.com/alirezarezvani/claude-skills.git synced 2026-08-28 04:24:58 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
claude-skills/agent-launcher/commands/cs-launch.md
Claude d1f2396c6f
feat(agent-launcher): new domain plugin for building Claude Managed Agents 
Adds the agent-launcher/ top-level domain — a plugin re-implementation of
Anthropic's launch-your-agent reference skill (Apache-2.0; independent, not a
fork) for building Claude Managed Agents (CMA) in the user's own account.

Every session starts with a goal (./my-agent/goal.json, surfaced by an opt-in
AGENT_LAUNCHER_SESSION=1 SessionStart hook + /cs:goal); loop_compiler.py
compiles that goal into a bounded grade->iterate loop (CMA user.define_outcome
self-grading, max_iterations 1..20), a recurring POSIX-cron scheduled-deployment
loop, or a single-pass interview->stage->launch workflow.

- 6 skills: agent-launcher-orchestrator (context: fork goal router) + interview
  + stage-launch + grade-iterate + run-without-you + wrap-up
- 18 stdlib-only deterministic scaffolder tools (NO network/API calls; live
  launches emitted as BYOK curl that never prints the key); all pass --help/--sample
- 4 agents (orchestrator + interviewer + grader + deployer), 8 /cs:* commands
- opt-in SessionStart/SessionEnd hooks (exit 0 on any error), 5 shared
  references, 4 assets (build-sheet schema + overview/next-directions templates
  + example)
- validators enforce CMA limits (<=20 skills/session, <=8 memory stores,
  depth-1 multiagent, max_iterations <=20, <=1000 deployments/org)
- registered in marketplace.json; headline counters trued up via
  derive_counters.py --check (skills 362->368, domains 18->19, tools 644->664,
  refs 741->746, agents 102->106, commands 116->124, plugins 88->89)

Distinct from engineering/agent-harness (generic bounded loop over any domain)
and engineering/write-a-skill (authors Claude Code skills, not CMAs).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012FwXG6TqCXKZQvF4iD69cv
2026-08-17 02:39:34 +00:00

1.4 KiB
Raw Blame History

description argument-hint
Main entry / resume for building a Claude Managed Agent. Runs the agent-launcher-orchestrator skill from the current session goal — reads ./my-agent/goal.json, routes to the right phase (interview → stage-launch → grade-iterate → run-without-you → wrap-up), compiles the loop/workflow, and forks to the phase skill. Emits BYOK curl; never makes API calls or prints the key. [optional: a one-sentence goal to set first]

/cs:launch — build/resume a Claude Managed Agent

Route through the agent-launcher-orchestrator skill.

$ARGUMENTS

Steps

  1. If $ARGUMENTS is a goal and no ./my-agent/goal.json exists, set it: python3 agent-launcher/skills/agent-launcher-orchestrator/scripts/goal_state.py init --goal "$ARGUMENTS".
  2. Route from the current phase: python3 agent-launcher/skills/agent-launcher-orchestrator/scripts/goal_router.py --out-dir ./my-agent — act on exit 0 (route) / 3 (ask the printed question) / 4 (refuse; get one sentence).
  3. Compile the loop: python3 agent-launcher/skills/agent-launcher-orchestrator/scripts/loop_compiler.py --out-dir ./my-agent.
  4. Invoke the routed phase skill; on completion, goal_state.py advance and print a ≤100-word digest (phase done, artifact paths, loop shape, one next step).

Refusals

  • No goal set → run /cs:goal set "..." first.
  • Under-3-word goal → get one sentence naming the one job.
  • Never touch the network or the API key.