cyber-aced

0.4.2

Patch Changes

• f017f7c: Resolve the contradictions a prompt audit found in the shipped skills. Eval results are read from the shared .agents/aced/results/<target-slug>/ directory everywhere, lowering a per-scenario threshold goes through the spec gate, name-only skills carry exactly By name only (the improve-skill Q3 check accepts it). init-aced registers architect-impl, the judge model defaults to the session's current model, the cyberplace version is resolved instead of pinned, and issue and ADR references are gone from the skill bodies.
• a35b43d: Align the impl-judge with the plugin contract (it now also loads gate-validation-governance), resolve sdd:automaton to sdd:sdd-automaton, use one cyberlegion@<version> form, name the gherkin-cli pin once, and drop history asides and repo-only references from agent and skill bodies.

0.4.1

Patch Changes

• 0c45202: define-skill and improve-skill now agree on how a skill description is written: say what the skill does, then when it applies, with no fixed lead-in such as "Use this skill when", and stay within 1024 characters.

  • The skill-design governance drops the rule that a description must contain "Use this skill when" or "When to use" and the 120-character cap.
  • Both skills read skill-design and agent-tool-output from a copy committed in the skill (after a .agents/governances/ override) instead of the stale cyberplace@0.2.4 CLI.
  • define-skill's audit step runs improve-skill's bundled validator in place of cyberplace audit validate, a command that never existed.
  • improve-skill's Q1 accepts a capability-first description with a "Use when" clause, warns on a fixed lead-in, and exempts "By name only".

0.4.0

Minor Changes

• 97b02d2: check-freshness now reads measured records. Run check-freshness --suite <suite> --arm <label> to find the newest record the bench engine wrote for that arm under .agents/aced/results/bench/<suite>/ and compare its evaluated set with the working tree. A changed tasks.json, a changed or removed check, or a changed file arm source reads stale; a record with nothing to compare reads absent. It exits zero only for current.

Patch Changes

• f15619c: improve-skill's validate engine no longer flags a description of exactly "By name only" as too short. That description marks a by-name skill on its own (ADR-0031), so a visible by-name skill such as bench passes Q2 without padding its description or hiding the command.
• a0f6c39: improve-skill's validate engine no longer crashes with EISDIR when a skill's scripts/ holds a subfolder such as scripts/vendor/. It walks the subfolders, so the Q11 and E9 script checks also cover nested scripts.

0.3.0

Minor Changes

• a6e6233: Add the measured layer's engine and its aced-bench bin. plan prices a run of a suite's tasks without spending; run --consent runs them for real in throwaway git worktrees through a headless Claude Code; compare tells a real change from noise with permutation tests and a regression verdict. An arm can be a git ref, a package version, or one swapped file. Suites live under .agents/aced/bench/<suite>/, records under .agents/aced/results/bench/<suite>/.
• 0bdbbd1: Add the bench skill: measures a change with the aced-bench engine. It checks a measured run fits the question, shows the plan and its price, runs only on an explicit yes to that plan (never with no person present), and reports the verdict without calling an unclear result safe or a price change a regression.
• bdb3c8b: Publish cyber-aced to npm so the aced-bench bin reaches consumers: a CI job or tool can run npx aced-bench without the ACED plugin installed. The bin ships compiled to dist/aced-bench.js, because Node does not strip types from files under node_modules. aced-bench help (or --help, -h) prints the usage and exits 0.

Patch Changes

• f4101cb: aced-fit gains a second axis, measured: worth | not-worth, for the coming measured layer: whether real headless runs answer a question simulation cannot. It is declared under bench: in eval.md and is opt-in, so an eval.md with no bench: key is not-worth and no existing subject changes classification.