v0.3.1 — self-healing db init (a chaos-test find)

Patch release. The randomized kill-schedule test in CI caught a real init crash window: a process SIGKILLed between schema creation and the version stamp left a root that failed every subsequent open with a duplicate-column migration error. Migrations are now idempotent and transactional — kill -9 anywhere during first start self-heals on the next open. Two regression tests pin the exact window.

If you created a root with v0.3.0 or earlier and it was interrupted during its very first start, v0.3.1 opens it cleanly.

(Yes, the chaos test we shipped two days ago paid for itself already — see CHANGELOG.md.)

v0.3.0 — the LLM-experience release: llm check / show / runs

Making the model behind agent.run operable like infrastructure — whether it's a CLI agent, a hosted API, or a model served on localhost. Schema v3 (auto-migrates).

Highlights

• forgeflow llm check — one live probe per agent binding and per embedding model: endpoint reachable, auth valid, model loaded, and the model can follow the fenced-JSON output contract. Exit 1 on failure — croneable. Replay bindings report available recordings.
• forgeflow llm show ROLE --data '{...}' — the exact assembled prompt a step would send, plus the prompt_sha the audit trail pins.
• forgeflow llm runs + per-model metrics — verdicts, wall-time, and re-ask counts per run and per model. Rising re-asks = your early warning that a model update started fumbling the contract.
• Three-stage binding validation, all fail-loud: structure at pack load, environment at engine start (CLI resolved and pinned, secrets present — after any --replay-from wrap, so replay CI never demands the live backend), live chain via llm check.
• params: passthrough for openai-compat (temperature, max_tokens, response_format — the Ollama/vLLM/llama.cpp knobs) and max_turns / extra_args for the agentic CLI backend.
• docs/LLM.md — copy-paste recipes for Ollama, vLLM, llama.cpp, LM Studio, gateways, the Claude CLI, and record/replay, plus the error-class troubleshooting table.

185 tests. Full details in CHANGELOG.md.

v0.2.0 — fan-out/join, versioning, schedules, HTTP dashboard, record/replay

The industrialization release. Schema v2 (auto-migrates existing roots).

Highlights

• Fan-out/join: fanout.emit runs a step per list item and fires a join event exactly once when every spawned task is terminal — crash/retry-safe barriers with truthful counts.
• Definition versioning: tasks are stamped with the workflow definition hash; editing YAML under a mid-flight task parks it for a clean re-run instead of replaying old state through a new graph.
• Timed triggers: schedule: fires events once per window — restart-safe cursor, no catch-up storms.
• HTTP front door: built-in dashboard + JSON API + POST /api/emit, bearer-token auth, loopback-safe defaults.
• Record/replay: every agent verdict is archived; --replay-from ROOT re-runs whole pipelines answering LLM steps from recordings — deterministic, token-free CI that fails loudly on prompt drift.
• Pack-tunable retry policy: retry: in project.yaml retunes error classes or defines new ones for pack block outcomes.
• Chaos-tested: randomized kill -9 schedules prove convergence and bounded side effects. 169 tests, stdlib only.

Also: MIT license, CI (Linux/macOS × Python 3.9–3.13), migration guides for Celery/cron/Airflow/LangGraph/Temporal in docs/COMING_FROM.md.

Full details in CHANGELOG.md.