TEAM.md (copy into the project repo)
This project is staffed from agentic-team (<repo URL>), installed in .team/.
- Profile:
<research | prototype | hobby | internal | production | other: reason>, set at intake (see.team/playbooks/engagement.md, Project profiles). Changes only on the owner’s message. - Team version:
<tag or commit>. Roles and playbooks at that version win over anything here about how agents work. - Agent tools in use:
<the agent tools this project uses>; installed with the generic adapter plus the matching tool adapter. State follows.team/docs/github-state.md. - Help pages:
<where users read how to use the product, e.g. docs/help/ or the in-app help screen; "none" for research>. The Product Manager owns their wording. - Spec of record:
<paths, e.g. docs/specs/v1.md, docs/specs/v1-openapi.yaml>; decision records indocs/decisions/. - Main branch:
<main>. Only the owner merges to it, except what the autonomy stage below allows. - Autonomy stage:
0while the gates are built, then1(owner merges everything). Moves to 2 or 3 only on the owner’s explicit message, linked here:<link>, after the entry criteria in the foundation playbook are met (override the numbers here if needed).CODEOWNERSis what enforces it.- Stage 2: copy, working notes and new tests auto-merge on green CI.
- Stage 3: also features and fixes inside owned paths auto-merge on green CI.
- The red zone, gate files and approved documents never auto-merge at any stage.
Limits (defaults; the Budget Analyst recalibrates from the ledger)
| Limit | Default |
|---|---|
| Tickets in progress at once | 2 |
| Fix rounds per ticket | 3 (trivial: 1) |
| Build sessions per ticket, by size | trivial 4 · S 5 · M 9 · L 13 |
| Spec review sessions per change spec | 2 (comprehension and readiness), counted apart from the build cap |
| Wall-clock per agent session | 60 minutes |
| Cleanup cycle | 2 weeks |
| Weekly status day | <day> |
Standing docs budget (DOCS_BUDGET) | 6,000 words across README, AGENTS, TEAM, DESIGN and docs/*.md |
Build sessions count prepare, build and verify sessions that end with a team comment: 1 prepare + (1 build + 1 verify) × (1 + fix rounds). Defaults allow S one fix round, M two plus two extra reviews, L three plus four; the third BLOCK still stops any ticket. Spec review and pricing sessions are counted on their own line, so a review never eats a ticket’s build budget. Size guide: S one module, no new interface, no UI; M a few modules or one new interface or screen; L several interfaces or screens, or a red-zone path. Anything bigger is split.
Hired
| Role | Persona | Hired for | Tool and model override (optional) |
|---|---|---|---|
| engagement-lead | Ella | always hired: the owner’s single point of contact and guide; can’t be removed | |
| product-manager | Priya | spec and dispatch | |
| architect | Arjun | build runs | |
| implementer | Ian | build runs | |
| verifier | Vera | build runs | |
| researcher | Rohan | <only if research is needed> | |
| critic | Quinn | <only if research is needed> | |
| evaluator | Max | <only if the product has prompts or models with a quality gate> | |
| budget-analyst | Bea | estimates, per-ticket caps, spend ledger | |
| ops | Otto | foundation, housekeeping, releases, production and incidents |
Delete rows for roles this project does not use.
Skills adopted for this project
Every role carries its team skills (.team/skills/, listed in each role file). Skills added for this project only, from the practice and skill review (.team/playbooks/engagement.md), approved with the brief and installed in a gate/skills PR:
| Skill | Used by | When | Source at commit | Licence | Scan report | Quinn’s verdict |
|---|---|---|---|---|---|---|
<name> | <roles> | <the task it's for> | <link at exact commit> | <licence> | research/skill-scans/<name>.json | <ADOPT, or ADOPT WITH NOTES: what the role ignores> |
Delete this section if the project adopted none.
Units
- A unit is one bounded change:
<e.g. one roadmap item>. - Branch:
team/<issue>-<slug>, one issue per unit (see.team/docs/github-state.md, Branches). - Default scope:
<paths a unit may touch> - Protected areas (need an explicit scope grant):
<e.g. migrations, auth, prod config, .github/, .env*> - Conventions:
<where code style, naming and test layout are documented>
Baseline
Commands that must pass at the base and at the head, from a clean state: <install, build, lint, unit tests, e2e>
Eval (omit if no AI component has a quality gate)
- Golden sets:
<path>; scorer:<command> - Gates:
<component: metric ≥ pass mark> - Caps and budgets:
<iterations, candidate rounds, API spend per stage> - Allowed mode for real data:
<e.g. no-training mode only>
Release (omit for research and prototype profiles)
See .team/playbooks/production.md, Release.
- Version scheme:
<semver; version file or latest tag>. Ops tags each production release after the watch window. - Preview deploy command (per PR, or the single staging slot):
<command> - Preview address pattern:
<url> - Deployed-version check:
<how to read the deployed revision> - Smoke checks:
<commands> - Production deploy:
<CI workflow name>, behind theproductionenvironment’s owner approval - Watch window:
30 minutes - Rollback:
<command>; backup restore:<command>
Operate (internal and production profiles)
See .team/playbooks/production.md. Monitoring services and their monthly cost (owner-approved): <uptime, errors, logs, backups, cost alerts>. Where user reports arrive: <email, form>.
Project-specific rules
Rules that only apply here (data, secrets, budgets, owner approvals). Keep them out of the role files.