Skip to content

version-1/tq-factory

v0.1.0

Run bounded coding work from Tasq issues with external agents

TQ Factory plugin

Japanese counterpart: README.ja.md.

TQ Factory is a coding-agent workflow plugin for projects that use Tasq's tq CLI as their work tracker. It coordinates bounded work through issues and comments while external agents or humans perform the work. It does not start Tasq's legacy orchestrator, run an agent daemon, or generate documents through an embedded model.

Install for local development

The package includes a portable root plugin.json, a Codex compatibility manifest, and a Claude manifest. To try it in a local Codex client, add this package to a local or repository marketplace and install it from that source; Codex supports $REPO_ROOT/.agents/plugins/marketplace.json for a repository marketplace. Then invoke tq-factory for a multi-issue coding goal. This package intentionally does not add a marketplace catalog or publish itself.

See the official plugin packaging guide for marketplace shape and local-client availability.

The target project needs:

  • a running issue-tracker reachable by tq;
  • a registered project key (tq project list); and
  • permission to read and write the target issues and comments.

Set TQ_API_URL or use --api-url for a tracker other than the local default. Before writing, verify the endpoint with tq --output json issue list --project <project-key>.

Minimal flow

  1. Ask the agent to use tq-factory for a concrete goal and project key.
  2. It creates or resumes a goal issue and child issues in Tasq.
  3. An external agent works on one assigned child in an isolated workspace and records progress and handoff evidence with tq comment add.
  4. An independent reviewer records verification. An authorized owner closes the verified issue and updates the goal index.

Read the skill and its linked workflow reference for exact command shapes and state rules.

Offline validation scenario

Review the scenario without calling tq. It checks that a user request is turned into safe issue-state and handoff decisions rather than a fictitious agent runtime.