Flowtend agent delivery governance
Flowtend is a project-scoped control plane for AI coding work. It keeps agent tasks,
validation evidence, Review packets, approval decisions, and automation API handoffs together
so engineering teams can ship agent-produced changes with a human gate.
Core product surfaces
- Project board for Backlog, In Progress, Validation, Review, and Done work.
- Review packet evidence for implementation notes, validation commands, repository links, and feedback.
- Project-scoped REST and CLI access for Codex, Claude Code, and custom runners.
- Audit exports, workflow summaries, wiki context, and operations metrics for delivery governance.
Flowtend Docs
This fallback exists for agents that fetch public docs without running JavaScript.
Use it like a small MkDocs index: open the raw markdown map first, then follow the deeper guide that matches the job.
Agent Entry Points
Route Contract
Public docs routes are under /docs. Live automation uses only project-scoped API routes under
/v1/projects/{project_id}/....
Flowtend Skill Authoring
This fallback exists for agents that fetch /docs/skill-authoring without running JavaScript.
Use the Flowtend public docs hub first, then use this page as the live skill-authoring contract.
Default Skill
Create or update the runtime-native Flowtend skill by default:
~/.codex/skills/flowtend-api/SKILL.md for Codex or
~/.claude/skills/flowtend-api/SKILL.md for Claude Code.
Use a CLI-only skill only when the user explicitly asks for repo-local or OSS-compatible commands.
Runtime Contract
- Use only project-scoped routes under
/v1/projects/{project_id}/....
- Check
GET /health before write operations.
- Use
FLOWTEND_API_BASE_URL, FLOWTEND_API_PROJECT_ID, and FLOWTEND_API_KEY.
- Do not echo secrets or write raw tokens into skill files, logs, or normal output.
- For backlog-clearing runs, work one actionable task at a time, refresh the task list after each completed task, and continue until no actionable backlog tasks remain.
Completion Policy
If repository review is enabled, finish repo work by pushing a branch and opening a PR. If review is disabled, finish by committing, reflecting on main, and pushing main. Keep verification inside implementation and validation guidance.