ticket-pipeline
Skill0 starsChanged 36 days agoApache-2.0View on GitHub
Use to run a whole ticket/prompt end-to-end, solo, as the orchestrator — intake+clarify with the operator, decompose and fan out to builder subagents, then reviewer → security-validator → assemble a PR → a free local-AI (Ornith) review pass with comments → out for external human review. The full ticket-to-PR lifecycle in one shadow session, operator kept in the loop only for questions. Builds on the `orchestrator` skill + the builder/reviewer/security-validator agent defs.
In this repository21 other files it gives its agents.
- browserbaseSkill
- cairnSkill
- developmentSkill
- dispatchSkill
- house-styleSkill
- kimi-subagentSkill
- lean-design-standardSkill
- mergeSkill
- nexus-authSkill
- nexus-jiraSkill
- orchestrateSkill
- orchestratorSkill
- planningSkill
- releaseSkill
- remote-opsSkill
- reviewSkill
- securitySkill
- specSkill
- vision-reviewSkill
- workflow-basicsSkill
- writing-skillsSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 35
Leave the PR open for human review, with the AI comments attached.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Ticket pipeline — Opus orchestrates ticket → PR, solo
You (Opus) drive the whole lifecycle. Workers are tiered subagents (.claude/agents/): builder (sonnet/medium), reviewer (sonnet/high), security-validator (sonnet/high). The operator is pulled in only to answer questions. This is the standalone-session version of the nexus dispatch pipeline — not a nexus run.
Input
A ticket or raw prompt. If it's a ledger ticket, read it (and close/update it at the end). Otherwise treat the prompt as the ticket.
The stages
1 — Intake + clarify (you/Opus) — the operator gate.
Read the ticket + the relevant code. Form the plan. Then split what's unresolved into facts vs decisions: a fact is answerable from the codebase/cluster/docs — look it up yourself, never ask the operator something grep can answer. Only decisions (trade-offs, scope edges, taste, spend) go to the operator. Surface all independent decisions at once via AskUserQuestion, each with your recommended answer first — operator answers once → you proceed uninterrupted. Exception — dependent decisions don't batch: when B only makes sense after A is answered (a design-tree walk), grill one question at a time down the tree, still recommending at each step. Only re-interrupt later if a builder hits a genuine blocker you can't resolve. Don't start building with unresolved must-know decisions.
2 — Decompose → build (builder subagents, parallel).
Split into independent, tightly-specced units (see orchestrator skill for the split rule + spec quality). Fan out builder subagents — multiple Agent(subagent_type="builder", …) calls in one message for parallelism. Each spec: inputs, expected output/artifact, acceptance criteria, conventions. Use worktree isolation if builders touch overlapping files (isolation:"worktree").
3 — Review (reviewer subagent). Feed the combined change to a reviewer subagent. Must-fix findings → loop back to a builder with a sharpened spec, then re-review. Don't advance with open must-fix items.
4 — Security validate (security-validator subagent). Once review is clean, run a security-validator subagent over the change. Real vulnerabilities → back to build → re-review + re-validate. Don't advance with unresolved high/critical.
5 — Assemble the PR (you/Opus).
Integrate the threads, resolve cross-unit conflicts, run the full build/test suite yourself. Branch off main, commit (per the repo's commit convention), push, open a GitHub PR via gh with a clear summary + what was verified. (cairn later — for now GitHub.)
6 — Local-AI review + comments (Ornith — a CALL-OUT, not a subagent).
Ornith can't be an Agent subagent (Anthropic-only). So: send the PR diff to Ornith — gh pr diff <n> piped to claude-ornith -p "review this diff, list concrete issues" (or the litellm API). Take its comments and post them on the PR (gh pr comment). This is the free sovereign first pass before a human spends time — flag Ornith's comments as such.
7 — External review (human). Leave the PR open for human review, with the AI comments attached. Report to the operator: PR link, what shipped, review+security summary, Ornith's notes. Done — the human decides merge.
Loop control
Findings flow back to you; you re-spec and re-delegate (build → re-review → re-validate) until review + security are clean, then assemble. You own the plan, the verification, and the integration; subagents own bounded execution.
Anti-patterns
- Skipping the intake gate → building on wrong assumptions. Ask everything first.
- Advancing past a red review/security stage → the gates are the point.
- Trust-merging subagent output unread → you verify; that's the orchestrator half.
- Serial builders for independent units → batch the Agent calls.
- Over-decomposing a small ticket → if it's one clear change, just build it (skip the fan-out), but still review + security + PR.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

