paqad-ai 1.6.0 → 1.7.0

package/CHANGELOG.md

@@ -1,5 +1,58 @@
 # paqad-ai
 
+## 1.7.0
+
+### Minor Changes
+
+- [#90](https://github.com/Eliyce/paqad-ai/pull/90) [`71711f0`](https://github.com/Eliyce/paqad-ai/commit/71711f0e6574654cf9826b03a5dcfb24ee3d5aaa) Thanks [@HLasani](https://github.com/HLasani)! - Add **rules-as-scripts** (issue [#89](https://github.com/Eliyce/paqad-ai/issues/89)): turn the prose rules under
+  `docs/instructions/rules/**` into deterministic, per-project verification
+  scripts that run as a sub-step of `feature-development.checks`, so rule
+  adherence no longer depends solely on the model remembering them.
+
+  All prompt-driven — no new user-facing CLI commands:
+  - `analyze rules` (`rule-analyzer`) — embeds stable `<!-- @rule RL-<hash> -->`
+    markers, classifies each rule `deterministic` / `heuristic` / `unverifiable`,
+    detects rules already enforced by ESLint/TS/etc., flags conflicts, and writes
+    a reviewable `docs/instructions/rules/rule-script-map.yml`.
+  - `generate rule scripts` (`rule-script-generator`) — authors one `.mjs` per
+    rule plus synthetic `__fixtures__/{pass,fail}`. A script that misclassifies
+    its own fixtures is rejected via the Decision Pause Contract — never
+    registered. Strict from generation; per-kind over-flag guard.
+  - `feature-development.checks.rule_compliance` runs the registered scripts
+    diff-scoped, hash-cached. `deterministic` findings block under `mode: strict`;
+    `heuristic` findings route to review and never block. Missing declared
+    binaries are reported and skipped, never crash the stage.
+  - `add rule` / `edit rule` / `remove rule` / `mark rule as unverifiable`
+    (`rule-editor`) — per-rule cascade with stable IDs; no global rebuild.
+  - `rule-script-reconciler` surfaces `RS-*` drift (rules edited without regen,
+    manual map edits, failing fixtures) at planning entry.
+  - New dashboard **Rule Compliance** card; onboarding plants the `analyze rules`
+    prompt. Engine exposed as the `paqad-ai/rule-scripts` subpath export.
+  - `finding-normalizer` promoted from the security capability to base so its
+    cross-capability vocabulary (`PEN-`/`DT-`/`MD-`/`RS-`) is no longer nested
+    under one capability.
+
+  Provider-neutral: the same prompt sequence and skills produce identical
+  artifacts across every supported adapter.
+
+## 1.6.1
+
+### Patch Changes
+
+- [#87](https://github.com/Eliyce/paqad-ai/pull/87) [`a5e350b`](https://github.com/Eliyce/paqad-ai/commit/a5e350b0d25b4cec7415d78467f624b4e664a1ca) Thanks [@HLasani](https://github.com/HLasani)! - Fix `design test` workflow scripts that failed on every onboarded project:
+  - `runtime/scripts/design/scan-tokens.sh` resolved its helper scanner via a
+    CWD-relative path, so Step 2 always exited 2 with
+    `error: skill scanner missing` when run from anywhere other than the
+    runtime root. Now resolves the helper relative to the script's own
+    location.
+  - `runtime/scripts/design/coverage.sh` aborted under macOS's stock Bash 3.2
+    (`set -u` + empty-array expansion) whenever the components directory or a
+    component's matching tests were empty — the common first-run case. Guarded
+    the three vulnerable `"${arr[@]}"` expansions with the
+    `${arr[@]+"${arr[@]}"}` idiom.
+
+  Closes [#86](https://github.com/Eliyce/paqad-ai/issues/86).
+
 ## 1.6.0
 
 ### Minor Changes