| name | agentic-build-runner |
| description | Execute backlog items from the SQL-first backlog through claim, implementation, tests, traceability updates, and commit-readiness, in any repo using the Agentic Development Platform. Use when asked to build, pick up, run, complete, verify, or orchestrate work from data/backlog.sqlite, especially with multi-agent backlog execution. |
Agentic Build Runner
Use this skill to turn backlog requirements into code, tests, and traceability
evidence.
Required Companion Skill
Load $agentic-traceability first. Its audit and append-only rules are mandatory.
Fast Slice Loop
Use a feature slice as the default unit in complete-dev mode:
- Plan: choose one bounded feature slice and the exact items/routes/contracts in scope.
- Design: decide behavior, RBAC/scope/state expectations, and test seams before editing.
- Build: edit the scoped code and directly related tests only.
- Test: run targeted checks first; reserve
npm run ci:governance for feature completion, governance/tooling changes, or pre-push.
Prefer one consolidated verification/audit event when one command set covers several
tasks, test cases, use cases, or success criteria. If a targeted check fails, record
it once with:
npm run backlog:fail -- --item S07-TASK-01 --summary "Targeted check failed" --evidence "npm run test:agent-context"
Build Loop
- Read
AGENTS.md.
- Snapshot workspace state:
git status --short --branch
- Generate a bounded context packet, then select and inspect one item:
npm run agent:next -- --workflow route --max-items 5
npm run backlog:show -- --item S07-TASK-01
- Confirm the item is tractable:
- feature dependencies and release band are understood;
- persona/RBAC workflow expectations are relevant to the implementation;
- routes, permissions, scope, and evidence paths are clear;
- tests and success criteria describe observable behavior.
- Claim the item before editing (claim a narrow write scope):
npm run backlog:claim -- --item S07-TASK-01 --actor agent --scope scripts/agent-context.mjs
npm run backlog:start -- --item S07-TASK-01 --summary "Started context broker work"
- Implement using normal repo conventions. Keep the write scope as narrow as possible.
- Run targeted checks first, then broader gates based on risk.
- Complete or block the item:
npm run backlog:complete -- --item S07-TASK-01 --summary "Implemented profile merge" --evidence scripts/agent-context.mjs
npm run backlog:verify -- --item S07-TASK-01 --summary "Verified" --evidence "npm run test:agent-context"
- At a feature/release checkpoint, or after process/tooling changes, run the governance gate:
npm run ci:governance
- Record a feature-level audit event with
$agentic-traceability before your final response.
Multi-Agent Orchestration
Use subagents only when the user explicitly asks for multi-agent, delegated, or
parallel agent work.
When authorized:
- keep the parent agent responsible for selecting and claiming the backlog item;
- delegate bounded side work only — code mapping, isolated implementation slices,
test review, or security/RBAC review;
- give workers disjoint write scopes (use
--scope on the claim);
- tell workers they are not alone in the codebase and must not revert others' changes;
- keep the SQL backlog, audit log, and final integration in the parent agent.
Do not let multiple workers independently mutate data/backlog.sqlite,
data/backlog-source.sql, or data/audit/audit-log.jsonl.
Done Criteria
An item is commit-ready only when:
- implementation matches the item title, tests, success criteria, and persona workflow;
- relevant negative tests cover scope/role/permission cases where applicable;
npm run backlog:verify has evidence;
- targeted checks from the context packet pass;
npm run ci:governance passes at the next feature/release checkpoint;
- a documented audit
decision explains any gate that was waived.