AgentStack
SKILL verified MIT Self-run

Opsman

skill-glapsfun-cnative-skills-opsman · by glapsfun

Local-first meta-agent orchestrator for Dev and Ops tasks. Use when the user asks to run opsman, orchestrate a multi-skill task, check opsman status, or wants a test-driven, evidence-gated execution loop that discovers local skills, selects a team, plans, implements, and validates command evidence. Triggers: opsman, orchestrate, meta-agent, orchestrator, run lifecycle, evidence-gated validation,…

No reviews yet
0 installs
5 views
0.0% view→install

Install

$ agentstack add skill-glapsfun-cnative-skills-opsman

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Opsman? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Opsman — Meta-Agent Orchestrator

Opsman converts a Dev or Ops request into a persistent, test-driven execution loop using repository-local skills and POSIX automation. Agents reason; the shell proves. All run state lives in the target repository under .opsman/ (gitignored), so a run started in Claude Code can be resumed in Codex.

The one rule

Interact with run state ONLY through the opsman kernel. Never edit files under .opsman/ by hand, and never skip recording an event. If you did work, opsman record it — unrecorded work does not exist.

The kernel lives at scripts/opsman inside this skill. Call it with the skill's absolute path, e.g. /scripts/opsman status.

Kernel verbs

| Verb | Purpose | | --- | --- | | opsman start "" | Build the skill registry, initialize a run (state DISCOVERING) | | opsman next | Render the context packet for the role that owns the current state | | opsman worktree [] | Create or verify the isolated run worktree | | opsman run-step | Execute one command-backed plan step under policy | | opsman validate | Run acceptance checks and capture evidence | | opsman status | Print the current run's STATE.md | | opsman record --event [--payload ] | The only way to change state | | opsman map | Rebuild .opsman/registry/ from discovered skills | | opsman validate-run [] | Check run artifacts for consistency |

Verbs judge, resume, and clean arrive in later milestones; the kernel rejects them with exit 2 until then. The UNDERSTANDING→VALIDATING phases enforce artifact and evidence gates: opsman record refuses phase-exit events until the required planning artifact, worktree, implementation evidence, or acceptance evidence exists and validates.

Lifecycle

States: DISCOVERING → UNDERSTANDING → SELECTING → PLANNING → TEST_DESIGN → IMPLEMENTING → VALIDATING → JUDGING → COMPLETED, with DIAGNOSING, REPLANNING, WAITING_APPROVAL, BLOCKED, ABANDONED on the side. The transition table is scripts/state-machine.tsv; an illegal event exits 3 and lists the legal events for the current state. See references/state-machine.md.

Failure handling

  • Exit 3: you sent an event the current state does not allow — read the

error, it lists legal events.

  • Exit 4: another opsman process holds the lock; do not delete

.opsman/lock unless the reported pid is dead.

  • Exit 5: artifacts are inconsistent — run opsman validate-run and report

findings to the user instead of hand-editing state.

  • Exit 7: a required tool (jq, git) is missing — tell the user.

See references/artifact-contract.md for the full file and exit-code contract, and references/safety-policy.md for risk classes R0–R4.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.