Documentation
Agent roles
The six built-in agent roles, what each reads, what each writes, and where its prompt template lives.
Source: src/agents/agent-schema.ts 6 roles
Built-in agent roles. Each is a configured row under `agents:` in project.yml - provider, prompt template, permission profile, and any skills.
planner
Reads the task brief and produces a plan.
-
reads: Task description, project rules, .vibestrate/roles/planner.md prompt, attached skills.
-
writes: Structured plan artifact.
-
prompt template:
src/agents/default-prompts/planner.md
architect
Expands the plan with implementation specifics.
-
reads: Plan + project rules + attached skills.
-
writes: Architecture artifact (module map, interfaces, data flow).
-
prompt template:
src/agents/default-prompts/architect.md
executor
Applies code changes in the run's worktree.
-
reads: Plan, architecture, review findings if any.
-
writes: File edits in the worktree; execution log.
-
prompt template:
src/agents/default-prompts/executor.md
fixer
Addresses review findings.
-
reads: Review findings, validation logs, current diff.
-
writes: Patches and a `finding-responses` artifact.
-
prompt template:
src/agents/default-prompts/fixer.md
reviewer
Critiques the diff against the plan.
-
reads: Plan, current diff, validation output.
-
writes: Findings + a review decision (APPROVED / CHANGES_REQUESTED / BLOCKED).
-
prompt template:
src/agents/default-prompts/reviewer.md
verifier
Final gate before merge.
-
reads: All artifacts from the run.
-
writes: Verification decision (PASSED / FAILED / NEEDS_HUMAN) + decision summary.
-
prompt template:
src/agents/default-prompts/verifier.md
