v0.2.0

[0.2.0] - 2026-07-14

Added

• shell.run built-in step — arbitrary shell commands with template placeholders, captured stdout/stderr, and optional declared output files
• examples/docs-pipeline/ — document stamping chain (shell → manifest rename → move)
• ordine --version prints the installed package version and exits 0
• README demo GIF (demo/demo.gif) recorded from demo/demo.tape

Changed

• README reframes Ordine as a universal watch → transform → deliver engine; image/CSV workflow remains the founding example
• README quickstart correctly describes six sample images in the built-in example

v0.1.0

[0.1.0] - 2026-07-14

Added

• Repo scaffold, tooling, CI (Step 1)
• Playbook schema, loader, JSON Schema export, and validation tests (Step 2)
• SQLite ledger with task state machine, branch attempts, flags, and name reservations (Step 3)
• Step contract, plugin registry, headless engine, and task workdirs (Step 4)
• Headless image steps (image.validate, image.white_to_alpha, image.trim, image.export) with ImageMagick/Pillow backends (Step 5)
• Folder watch and manual triggers with settle detection, startup rescan, and ledger sink (Step 6)
• Ledger.create_task_arrival — atomic arrival-order ordinal assignment + insert (planned Step 3 amendment, landed Step 6)
• Pipeline runner with retries, recovery branches, flag escalation, manifest naming, and PipelineService (Step 7)
• file.rename_from_manifest and file.move built-in steps; LedgerNamingService; job manifest parsing (CSV/JSON/txt)
• Typer CLI (ordine) with XDG app config, --json output contract, and run --oneshot (Step 8)
• Ledger read helpers for CLI (list_pipelines, find_pipeline_id, list_branch_attempts, list_open_flags)
• StepRegistry.list_step_metadata() for plugin visibility
• FastAPI web UI (ordine serve) with ServiceManager, dashboard, tasks, flags, settings, and vendored HTMX 2.0.4 (Step 9)
• Pipeline editor with form ⟷ YAML conversion, immutable version history, diffs, revert, and branch-from-version (Step 10)
• dump_playbook() in core with round-trip guarantee; docs/editor.md
• Form-tab recovery branch editing (step-level and pipeline-level) with HTMX add/remove fragments (Step 10 follow-up)
• Semantic version diffs, metadata-only diff notice, and editor form labeling polish (Step 10 follow-up 2)
• Structured change summary and side-by-side diff view (web/diffing.py; Step 10 follow-up 3)
• Dry-run lab with step-through, sandbox isolation, checkpoint resume, and ordine dry-run CLI (Step 11)
• Provider-agnostic LLM connector: raw REST adapters (Anthropic/OpenAI/compatible), keyring keys, JSONL audit log, token budget, settings UI, and ordine llm check (Step 12)
• AI drafting, failure diagnosis, and learned recovery branches with explicit approval only (Step 13)
• Manifest trigger (ManifestTriggerService) with row ordinals, mrow dedup keys, and reservation-at-creation sink (Step 14)
• llm.generate_image step with mock and OpenAI image providers, ImageBudget, and session_image_cap config (Step 14)
• Chained example under examples/chain/ and docs/chaining.md (Step 14)
• Workdir retention cleanup (ordine cleanup, [retention] config, Ledger.clear_workdir) (Step 15)
• ordine example quickstart scaffolder with CI-guaranteed oneshot test (Step 15)
• Release tooling: scripts/bump_version.py, scripts/build_deb.sh, release.yml, deb-smoke CI, version sync test (Step 15)
• Docs index, install/security/release guides, community files, demo/demo.tape, naming checklist (Step 15)

Changed

• Product renamed to Ordine: PyPI/deb/CLI/import package ordine, config/data dirs, keyring service, entry-point groups ordine.steps/ordine.engines, CI Python matrix via uv python install (Step 15 naming)

• Pre-public polish completed for service transitions, runtime checks under python -O, logging, and duplicate cleanup

• Repository identity transferred to Antikatoptis-Pareidolia/ordine with final license attribution and release URLs

• Final audit hardening added output-name containment, full-origin POST checks, HTMX restrictions, secret redaction, first-contact errors, documentation truth fixes, packaging assertions, and per-core-module coverage floors

• Pillow white-to-alpha uses get_flattened_data(); diagnosis repair calls use repair_diagnose purpose tag (Step 15)

• Flags inbox shows one-line hints per known flag kind (Step 15)

• Step 14 follow-up: manifest build_trigger_service rejects caller sinks; manifest_sink caches rows by mtime; trigger/chaining docs corrected

• Chain example uses on_collision: replace for regeneration; ordinal_regex skips non-matching files with explicit warning

• Mock image provider varies PNG bytes with prompt (sha256 fill + prompt text); deterministic for same ordinal/prompt/size

• Mock image provider keeps a white background with a prompt-keyed accent band so chain cleanup rehearsals exercise white_to_alpha/trim

• Test convention rule 21: no patching the subject under assertion; monkeypatch reserved for externals or delegating counters

• Step 13 follow-up: web wiring scope convention, full-stack draft MockTransport test, CI llm_live filter

Fixed

• Deb build uses fpm --deb-recommends (not invalid --recommends); deb-smoke CI logs fpm --version (Step 15 follow-up)

• CI deb-smoke container steps use shell: bash so set -euo pipefail works (Step 15 follow-up 3)

• /usr/bin/ordine deb symlink uses absolute /opt/ordine/bin/ordine; build asserts package contents (Step 15 follow-up 4)

• Deb build/CI assertions capture producer output before grep -q/head to avoid SIGPIPE under pipefail (Step 15 follow-up 5)

• Deb venv uses --copies (real interpreter binary); removed self-referential python3 symlink (Step 15 follow-up 6)

• Sigint integration test prints full subprocess stderr on failure (Step 15 follow-up 6)

• Lab dry-run ordinals now mirror production trigger semantics (no false pass for ordinal-dependent playbooks)

• Drafting prompts instruct models to configure ordinal sources when manifest/numbered files are implied

• Dry-run and lab runner-equivalent paths now report real step failure messages (not "no attempts executed") when the primary attempt fails before branch retries

• Flag escalation groups primary attempts by failing step (last_step_id) so multi-step tasks raise level 1+ flags correctly

• Ladder-scoped flag escalation counts only the failing step's primary and branch groups; unfinished attempt rows are ignored

• Recovery branch names must be unique across the entire playbook (not only within one on_failure block)

• Dashboard start/pause shows pending disabled state; paused pipelines display "last ran" version wording

• Startup rescan seeds the settle tracker instead of emitting partial files (Step 6 follow-up)