TFactory

Teach an AI coding assistant how to work in the TFactory codebase — the autonomous test-generation + execution platform. Covers the 4-agent pipeline, the LLM-provider abstraction (never call anthropic directly), the workspace layout, branching model, how to run the suite, and the non-negotiable conventions (i18n, no-automatic-pushes, dry-run side-effects).

Close the AIFactory ↔ TFactory loop hands-off. After TFactory tests a feature and finds problems, this drives one bounded cycle — hand the correction back to AIFactory, wait for its QA Fixer, re-test in TFactory — and stops when the run passes, the same tests keep failing (no progress), or a correction-cycle cap is hit (→ stuck for a human). One cycle per invocation; drive it with /loop for a fully autonomous test → fix → re-test loop. Epic

When TFactory's tests find problems in a feature, hand a correction back to AIFactory for revision. Reads the correction request TFactory's Triager already prepared (findings/handback_request.{md,json}), previews the target AIFactory spec, and on confirmation sends it so AIFactory's QA Fixer writes the fix on the original spec. The reverse direction of /handover-to-tfactory — together they close the AIFactory ↔ TFactory loop (epic

Hand a finished feature off to TFactory for autonomous polyglot test generation — from AIFactory, Claude Code, or any tool (via the MCP control plane, or a markdown/Gherkin/EARS acceptance-criteria file ingested with spec_sources.py). Records the task, snapshots the spec, and drives the full Planner → Gen-Functional → Executor → Evaluator → Triager pipeline across the v0.2 5-lane spine (unit / browser / api / integration / mutation) to produce a triage report + (optionally) commit tests to the feature branch + post a PR comment. Pair with /tfactory-watch under /loop to poll the task home and verify coverage — the full round-trip.

From an AIFactory project, receive a correction hand-back from TFactory (sister project). When TFactory's tests found problems, this previews the prepared correction and applies it to the original spec via AIFactory's QA Fixer (writes QA_FIX_REQUEST.md + runs the fixer). The reverse of /handover-to-tfactory.

From an AIFactory project, hand a finished spec off to TFactory (sister project) for autonomous test generation. Records the task with TFactory's MCP server; once TFactory Tasks 5-8 land, this also drives the planner→generator→executor→evaluator→triager pipeline.

Run a read-only cloud infrastructure assessment — discover an AWS/Azure/GCP account's resources, detect misconfigurations with Prowler (CIS/OCSF), build a service-topology diagram, and emit a verdict (reject/flag/accept). Follows the access → discover → diagram → assess flow and writes the report into findings/. Use for cloud security posture, misconfiguration audits, or "what's in this account".

Draft, preview, and publish a post to the TFactory GitHub Pages blog (docs/_posts/). Writes a well-structured Jekyll post in the TFactory voice, drops it into the Gruvbox-themed Pages site, previews it with the nix-Chrome screenshot harness, and (on request) commits + syncs dev→main. Use to write or refresh TFactory blog posts.

Pick one of the 15 v0.2 starter templates (5 each for pytest / Jest / Playwright), fill its vars, and drop the result into the project as a ready-to-run test file.

Produce a realistic, multi-pane screencast demo of the Claude Code → TFactory handover workflow, end to end and the same quality every time. Picks a known-good scenario (greeting-generator / python-unit / multi-lane / failure-flow), seeds its spec, starts the portal + backend, hands the feature off to TFactory, runs the full Planner → Gen-Functional → Executor → Evaluator → Triager pipeline, records the Claude Code terminal (asciinema) and the TFactory portal (Playwright) in lockstep, stitches them with the triage report into one composite demo.mp4 + demo.gif, and refuses to publish until an automated quality gate passes. Use to make demos.

Poll a running TFactory task to completion, then pick up its triage report and verify the generated tests cover the acceptance criteria. One check per invocation — drive it with /loop for a hands-off round-trip. Reads task_status over MCP and the triage report file directly (no backend dependency on the bugged report_get).

Add ONE generated test to an existing project by running the Gen-Functional sub-flow locally, without the full Planner → Executor → Evaluator → Triager pipeline.

Scaffold a .tfactory.yml + empty .tfactory/tests-catalog.json in an AIFactory repo so TFactory can start generating tests for it.