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)