v0.1.2

Changed

• Pack hone's review into units of consecutive tasks up to ~400 changed lines instead of one unit per task, and after round 1 re-review only units not yet done: a unit is done after two consecutive clean reviews, and a fix re-opens every unit whose files it touches. Each round still logs a feature - Rating: line (the lowest unit rating), which smithy reads; the worked example's code-review.md shows the units (#35)
• Drop "Don't do this" items and sentences that restate a rule already given in its own section of the same skill (forge, anvil, temper, quench, hone), and merge temper's and hone's Convergence Rule and Iteration Cap sections. No rule is removed; hone is back within the 260-line prompt budget, so ./install.sh eval scores 100 (#31)

Added

• Add --runs N and --critic-timeout to ./install.sh eval --live-llm: repeat the corpus and report recall per run, its mean and range, clean-control findings per run and how often each defect was found; and give slow backends more than the fixed 170 s per critic call (#34)
• Document the scorer-tuning rule in evals/program.md: diagnosis terms added after reading replies come from a backend other than the one being reported, and the reported run uses a frozen scorer (#34)
• Add read-only review agents temper-critic (Read, Glob, Grep), hone-critic (plus Bash, for repros run in a throwaway worktree) and review-judge. Temper and hone now write each round's inputs to a bundle file and dispatch these agents on it, so critics can't edit the work under review and stay blind to earlier rounds and ratings. tools: is enforced only by harnesses that honour it; elsewhere the agents' no-edit rule still binds (#33)

Fixed

• Stop install.sh looping forever when a link target and its directory share no root (e.g. two drives under Git Bash): relpath now fails, and the link is skipped with an error and counted as a problem (#32)
• Check install.sh's skill lists in CI with an in-process loop instead of printf | grep -q, which can die of SIGPIPE once pipefail is on (#32)

v0.1.1

Changed

• Rewrite the README around a quick start: the real clone URL, how to start a feature with smithy, red-amber-green next to the gates it belongs to, a table of the damascus commands, and a vX.Y.Z upgrade example instead of a release that doesn't exist (#28)
• Mark docs/production-readiness.md as a record of the original plan, with master protected by a repository ruleset, and describe the docs as harness-agnostic rather than Claude Code-only (#28)

Added

• Add docs/reference.md, a configuration and CLI reference written from the code: model routing through .damascus/models.conf and DAMASCUS_* variables, every gate's flags, exit codes and file formats, stage addenda, and the evaluation harness with the live critic benchmark (#28)
• Add evals/run_eval.py --check-corpus, run in CI, which fails when echoing the prompt earns credit, when a defect is not matched by its own description or is matched by a sibling's, or when a NIT counts as BLOCKING (#27)

Fixed

• Make upgrade-consumers run each repo's own vendor/damascus/install.sh, so its links keep pointing at its pinned submodule instead of the checkout running the command; skip a repo without one unless --link-to-this-checkout is given, and exit 1 when an install or verify fails (#29)
• Cap a review round at B+ in the eval's reference adjudicator at under 25 % overlap, as the temper and hone skills specify, instead of 20 % (#29)
• Stop damascus agent-run from falling back to a hardcoded Ollama model name; with no model pulled, configured or in OLLAMA_MODEL, it now says so and exits 1 (#29)
• Correct the test-lock manifest name in its docstring to .damascus/locks/amber-tests.sha256.json (#29)
• Fix the README's install and upgrade snippets, which committed whatever master was instead of the pinned tag: initialise damascus's submodules from inside vendor/damascus rather than from the repo root (#28)
• Make ./install.sh eval --live-llm measure something: a seeded defect now counts only when a [BLOCKING] finding names it and diagnoses it in words absent from the prompt. Before, any keyword anywhere in the reply counted, and echoing the prompt scored 12/12. The clean control is now scored for false positives, the echo baseline is reported, and every prompt and raw reply is saved under .damascus/eval-runs/. The corpus keywords field becomes anchors + diagnosis (corpus 0.3.0) (#27)
• Fix the clean control's FR-001, which did not say how to round, so INV-001 failed for sub-cent subtotals; it now rounds down (#27)

v0.1.0

Initial release.

Added

• Add the five SPDD stage skills forge, anvil, temper, quench and hone, and the smithy orchestrator that resumes any feature from its disk artifacts (#1)
• Add install.sh to symlink the skills, aliases, agents and KEEP-class superpowers skills into a consumer's .claude/ with relative links (#1, #3)
• Support stock macOS bash 3.2 in install.sh (#3)
• Add --verify, --dry-run and --uninstall to install.sh, which touches only links it owns and exits non-zero when a link cannot be placed (#3, #19)
• Vendor obra/superpowers 7.0.0 and github/spec-kit 1.2.0 as submodules pinned to upstream tags (#18, #26) (e17bbff)
• Add the DENY / KEEP / CONDITIONAL policy that keeps seven superpowers skills out of the install, including the transcript analyser diagnosing-superpowers (#1, #26) (e17bbff)
• Add the quench dispatch agents bdd-scenario-writer, tdd-test-generator, playwright-e2e-tester, fastapi-implementer and labcoat (#1) (cc17dc9)
• Add the aliases prd-authoring, speckit-decomposition, adversarial-review-loop, bdd-tdd-execution, code-review-loop and spdd-pipeline (#1) (cc17dc9)
• Add red-amber-green to quench: tests freeze at amber, then mutation, static, traceability and stable-green gates (cc17dc9)
• Add temper's mandatory critic procedures and overlap signal; A++ takes two consecutive clean rounds (cc17dc9)
• Add hone, which reviews the diff in units of at most 400 lines and shows every fix failing without itself (#12) (cc17dc9)
• Add smithy's finish step, which updates README and CHANGELOG, runs Atlas when installed and logs FINISH: (ebfc01e)
• Add stopping rules for long runs: keep going between gates; stop at a gate, a refusal or before anything destructive (#22)
• Lead every gate message with a "Needs your call" block (#22)
• Require a repro or an implementer's question on blocking findings, and add the UNCONFIRMED finding class (#22)
• Run quench's [P] tasks in parallel git worktrees and merge them one at a time (#22)
• Add examples/cart-discounts/, one feature's artifact trail through all five gates (e17bbff)
• Add damascus agent-run, a stdlib-only runner that sends a critic, judge or implementer prompt to a local OpenAI-compatible server (llama.cpp, vLLM, LM Studio, SGLang, LocalAI), Ollama or an agent CLI (claude, gemini, codex, opencode, llm), with --probe, .damascus/models.conf routing and model/provider self-identification stripped for blind judging (#24)
• Add mechanized gates as install.sh subcommands: revert-proof (each fix reverted in an isolated worktree must turn its test RED, not GREEN or INCONCLUSIVE), proof-coverage (every changed production file has such a proof), test-lock (amber tests frozen by SHA-256, verified at green) and bdd-check (every .feature file is bound to a step runner) (#24)
• Add install.sh upgrade-consumers to inspect or upgrade downstream repos' damascus installs, skipping read-only mounts (#24)
• Add ./install.sh eval (evals/run_eval.py), a scored harness over the smoke test, the mechanized gates, skill invariants and a prompt-size budget, with a seeded-defect corpus and an autoresearch-style evals/program.md (#24)
• Add --all-harnesses to also link skills and agents into .agents/ and .gemini/ (linked automatically when those directories exist), and --print-root (#24)
• Add .damascus/addenda/<stage>.md hooks so a host repo can add domain checks or a fourth critic lens to any stage without forking it (#24)
• Add [DELTA] / [INVARIANT] requirement tags and INV-NNN safeguards to anvil, a vacuity check for every SC in temper's testability lens, and a one-round M0 micro-temper for localized spec changes (SCR-NNN) (#24)
• Cap a review round at B+ for low overlap only when at least four distinct blocking findings were raised and under 25 % of them by two or more critics, so disjoint nits can no longer block convergence (#24)
• Keep critics blind to the round number and prior grade, pass only the previous round's diff after round 1, and rotate a critic on the A+ → A++ confirmation round (#24)
• Add a bounded autoresearch keep/revert loop to quench for [HARD] tasks with a quantitative target (#24)
• Switch the dispatch agents to model: inherit, cut each to at most 150 lines, and teach Pydantic v2, SQLAlchemy 2.0, timezone-aware UTC and property-based tests for INV-NNN safeguards (#24)
• Add references/, the research notes behind these changes, and tests/dogfood_e2e.sh, an end-to-end run of every gate on a throwaway project (#24)
• Document the pipeline in a Sphinx site with mermaid architecture diagrams, generated from the shipped files (#23)
• Document install, upgrades, pins and the superpowers policy in the README (#1) (e17bbff)
• Document the single-founder production-readiness plan (#3)
• License the project under MIT; vendored submodules keep their upstream licenses (#3)