using-agent-dev-team
Skill9 starsChanged 2 months agoMITView on GitHub
Routes an incoming engineering request to the right team member and the right workflow, and explains the tier system. Use at the start of any development work when it is unclear who should do it, which skill applies, or how much process a task deserves. Also use when a task turns out to be bigger than it looked and needs re-routing.
In this repository28 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- api-designSkill
- architecture-decisionSkill
- autonomous-relaySkill
- browser-verificationSkill
- code-review-passSkill
- data-analysisSkill
- data-modelingSkill
- data-pipelineSkill
- data-qualitySkill
- documentationSkill
- firmware-buildSkill
- frontend-buildSkill
- incident-responseSkill
- incremental-deliverySkill
- ml-lifecycleSkill
- performance-passSkill
- release-and-rollbackSkill
- requirements-interviewSkill
- security-hardeningSkill
- simplification-passSkill
- spec-writingSkill
- systematic-debuggingSkill
- tdd-loopSkill
- team-escalationSkill
- work-breakdownSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 26
When two classes fit, take the heavier one.
- Distinctive: line 81
Follow the skill. If it has a checklist, track it.
- Distinctive: line 82
Verify with evidence. Quote what you ran and what it printed.
- Very specific: line 83
If you hit a ceiling, stop and use team-escalation.
- Very specific: line 94
If the harness exposes a plugin root variable, references/ sits directly under it.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Using the Agent Dev Team
This is the entry point. It answers three questions: how much process does this deserve, who should do it, and which workflow applies.
Step 1: Classify the work
| Class | Looks like | Process |
| Trivial | A typo, a constant, a rename with known call sites | Do it. No ceremony. Still run the tests. |
| Bounded | A change to a flow that already exists in this repository | Clarify, short plan in chat, implement, verify |
| Feature | New behavior, several files, requirements not fully written | Requirements, breakdown, then build task by task |
| Architectural | New subsystem, changes a contract others depend on, hard to reverse | Requirements, ADR, breakdown, then build |
| Incident | Production is broken right now | Stop. Go to incident-response. |
When two classes fit, take the heavier one. The ratchet is one way: discovering hidden complexity upgrades the class mid-task. Nothing downgrades it.
Step 2: Route to a team member
| Situation | Agent |
| Fuzzy ask, no written requirements | product-manager |
| Requirements exist, no plan | tech-lead |
| Fully specified, mechanical, at most two files | intern-engineer |
| Clear requirement, existing pattern to follow | software-engineer |
| Ambiguous, cross-cutting, or a migration | senior-engineer |
| Two designs disagree, or a subtle correctness bug | principal-engineer |
| A change is written and needs a gate | code-reviewer, then test-engineer |
| Touches auth, secrets, untrusted input, crypto | security-auditor |
| Measured slowness or a missed budget | performance-engineer |
| Deploy, rollback, alerting, runbook | sre |
| Production broken right now | incident-commander |
| MCU, RTOS, driver, memory budget | firmware-engineer |
| New board, boot failure, signal-level bug | board-bringup-engineer |
| Component, state, layout, Core Web Vitals | frontend-engineer |
| Usability, accessibility, interface copy | ux-reviewer |
| README, ADR, runbook, changelog | docs-engineer |
Route to the lowest tier that clears the ceiling. Over-assignment is not caution; it hides the fact that the brief is too vague, and it costs more.
Step 3: Pick the workflow
| Phase | Skill | When |
| Define | requirements-interview | The ask is fuzzy |
| Define | spec-writing | Requirements exist and need a written spec |
| Plan | work-breakdown | A spec needs splitting into verifiable tasks |
| Plan | architecture-decision | A non-obvious technical choice needs recording |
| Plan | api-design | An interface others will depend on is being designed |
| Build | tdd-loop | Any implementation |
| Build | incremental-delivery | Executing a multi-task plan |
| Build | frontend-build | The change is user interface |
| Build | firmware-build | The change runs on a microcontroller |
| Verify | systematic-debugging | Something is broken and the cause is unknown |
| Verify | browser-verification | A UI change needs real-browser confirmation |
| Review | code-review-pass | A diff needs a gate |
| Review | simplification-pass | Code is more complex than the problem |
| Review | security-hardening | A trust boundary is involved |
| Review | performance-pass | Something is measurably slow |
| Ship | release-and-rollback | Getting it into production |
| Ship | incident-response | Production is broken |
| Ship | documentation | Docs are missing, stale, or wrong |
| Meta | team-escalation | A task exceeded the current tier's ceiling |
Step 4: Work
- Announce the class, the agent, and the skill in one line, so the human can override it before work starts.
- Follow the skill. If it has a checklist, track it.
- Verify with evidence. Quote what you ran and what it printed.
- If you hit a ceiling, stop and use
team-escalation. Escalating with a handoff packet is the system working, not a failure.
Finding the reference files
Skills and roles cite checklists as references/<name>.md. That path is relative to the
agent-dev-team installation, not to the project you are working in, so a plain read
from a project directory will miss.
Resolve it once, at the start of a session that needs one, and reuse the answer:
- If the harness exposes a plugin root variable,
references/sits directly under it. - Otherwise search for
**/references/escalation-ladder.mdacross the installation directory —~/.claude/plugins/,~/.claude/skills/, or wherever the repository was cloned. - Under a copy-mode install the directory is
~/.claude/skills/adt-references/.
A checklist that cannot be found is not a reason to skip it. Say you could not load it and work from the process in the skill body, which is self-contained.
Red flags
| Thought | Reality |
| "I know what to do, I'll skip routing" | Routing takes ten seconds and prevents the wrong tier doing the work. |
| "I could not find the checklist, so I skipped that step" | Say so. The skill body carries the process; the checklist adds depth. |
| "I'll assign it senior to be safe" | Over-assignment hides a vague brief. Fix the brief. |
| "It grew but I'm almost done" | Re-classify. Hidden complexity upgrades the class mid-task. |
| "No skill fits exactly" | Take the nearest. Partial process beats none. |
| "The human is in a hurry, skip the gates" | Urgency is when gates pay. Say what you skipped, if you must skip. |
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.