Skip to content
Agentmart
Submit a tool

orchestrate-sentinel

Manage and execute Sentinel CLI, observability-gateway, test-platform, installer, chart, and docs work through the repository's central .orchestrator run graph.

Skill by caglarkc

IndexedMIT★ 2
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/caglarkc-orchestrate-sentinel

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Manage and execute Sentinel CLI, observability-gateway, test-platform, installer, chart, and docs work through the repository's central .orchestrator run graph. Use for project planning, PM decomposition, architecture/contracts, implementation, review, verification, integration, resuming prior work, or coordinating Codex/Cursor/Claude agents. Enforces gateway read-only, install-path separation, evidence contracts, risk gates, and Git checkpoints.

Use cases: testing, data, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/orchestrate-sentinel

What's inside

Skill name
orchestrate-sentinel
Description
Manage and execute Sentinel CLI, observability-gateway, test-platform, installer, chart, and docs work through the repository's central .orchestrator run graph. Use for project planning, PM decomposition, architecture/contracts, implementation, review, verification, integration, resuming prior work, or coordinating Codex/Cursor/Claude agents. Enforces gateway read-only, install-path separation, evidence contracts, risk gates, and Git checkpoints.
License
MIT
Instructions
91 lines in SKILL.md · view on GitHub

Agent Skills spec issues

  • files: referenced but missing: scripts/auto-push-watch.sh

Files (4)

  • SKILL.md
  • openai.yaml · agents/
  • module-routing.md · references/
  • run-patterns.md · references/

SKILL.md (first lines)

# Orchestrate Sentinel

Use `.orchestrator` as the source of truth for non-trivial project work. Do not manage multi-step implementation only in conversation memory.

## Start

1. Read repository-root `README.md` and `INSTALL.md`.
2. Read `.orchestrator/PROJECT-STATE.md` and `.orchestrator/SYSTEM.md`.
3. Read `.orchestrator/COMMIT_CONVENTION.md` and the assigned role under `.orchestrator/roles/`.
4. Run `node .orchestrator/bin/orchestrator.mjs discover` when catalog is missing or repository structure changed.
5. Inspect active runs before creating a duplicate run.

For module and capability selection, read [module-routing.md](references/module-routing.md). For reusable graph shapes, read [run-patterns.md](references/run-patterns.md).

## Choose workflow

### Answer, audit, or design only

Use a read-only `analysis`, `specification`, `architecture-review`, or `pm-planning` item. Do not infer implementation authorization.

### Build or change

Create or resume a run. Apply contract-first ordering:

```text
PM scope -> architecture/contract -> implement -> review -> verify -> integration -> PM acceptance
```

Add independent review and verification for every item required by `.orchestrator/config.json` risk policy.

### Diagnose

Create read-only discovery/reproduction items first. Do not add a fix item unless the request includes fixing or the user later authorizes it.

### Resume

Run `validate`, `status`, then inspect `events.jsonl` and `results/`. Reconcile stale `active` ownership before redispatch.

## PM Manager protocol

- Preserve the complete approved scope; never collapse Compose, Kubernetes, and COS into one installer story.
- Keep the observability gateway read-only unless the user explicitly approves `gateway-write-expansion`.
- Treat `agentic/` as reference code, not product source.
- Assign role, capabilities, write scopes, risks, approvals, outputs, and exact acceptance to every item.
- Keep architecture, implementation, review, verification, and integration ownership separate.
- Record durable decisions in `run.decisions`, not only prose.
- Create revision items; never rewrite failed history.
- Do not mark completion until accepted integration and documentation synchronization exist.

## Dispatch protocol

1. Run `sync` and `status`.
2. Select the first safe batch.
3. Prefer native subagents for bounded independent items when available.
4. Use worktrees for parallel writers when supported; otherwise serialize.
5. Render a platform handoff when native delegation is unavailable:

```bash
node .orchestrator/bin/orchestrator.mjs render <run.json> <item-id> --platform cursor
```

6. Require `.orchestrator/contracts/result.schema.json` output.
7. Record accepted attempt with `record`; do not manually set `done`.
8. After a write item passes its checks, record one atomic Conventional Commit with `type(scope): summary`. `Work-Item` and `Phase` footers are optional.
9. If the commit message is valid, `git push` is allowed. Never force-push and never use `scripts/auto-push-watch.sh`.

## Role routing

- `pm-planning`, backlog, scope, acceptance: `roles/pm-manager.md`
- architecture, ADR, schema, module boundary: `roles/architecture-manager.md`
- code/config/test/docs mutation: `roles/code-implementer.md`
- independent diff/contract review: `roles/independent-reviewer.md`
- test execution: `roles/verifier.md`
- secret/token/approval/gateway-auth: `roles/security-data-reviewer.md`
- accepted cross-module assembly: `roles/integration-manager.md`

## Non-negotiable product invariants

- Gateway stays read-only; no alert, dashboard, backend write, or Grafana proxy without approval.
- CLI talks to Prometheus/Loki/Tempo only through the gateway.
- Secrets stay in environment variables, not committed YAML.
- Compose, Kubernetes, and COS install paths stay distinct; COS installer is not complete.
- `agentic/` is reference, not the shipped product.
- Tool approval, timeouts, and redaction are not silently disabled.
- Do not claim a check ran when it did not.

## Completion

Return the outcome, accepted artifacts, executed checks, commit SHA, remaining risks, and next graph state. Read-only items do not create empty commits. Valid Conventional Commit messages may be pushed with `git push`; never force-push.

From caglarkc/sentinel-coming, MIT.

More from caglarkc/sentinel-coming

Signals

Stars
2
Forks
2
Open issues
0
Last commit (this folder)
Aug 15, 2026 (2 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from caglarkc

Cite this page

orchestrate-sentinel skill. Agent Mart. https://agentmart.dev/skills/caglarkc-orchestrate-sentinel (last updated 2026-10-24)