paqad-ai 1.3.0 → 1.5.0

package/CHANGELOG.md

@@ -1,5 +1,35 @@
 # paqad-ai
 
+## 1.5.0
+
+### Minor Changes
+
+- [#82](https://github.com/Eliyce/paqad-ai/pull/82) [`f330595`](https://github.com/Eliyce/paqad-ai/commit/f330595d38c5e66afc54443daefe9e4f97a7ea68) Thanks [@HLasani](https://github.com/HLasani)! - Living module lifecycle: prospective decisions, retrospective reconciliation, test-driven health, dashboard drift checks (closes [#80](https://github.com/Eliyce/paqad-ai/issues/80))
+
+## 1.4.0
+
+### Minor Changes
+
+- [#75](https://github.com/Eliyce/paqad-ai/pull/75) [`283d973`](https://github.com/Eliyce/paqad-ai/commit/283d9736621e7571013a377408a8ec6416b54df9) Thanks [@HLasani](https://github.com/HLasani)! - Gate every agent turn on the entry-file load, not just code-mutating tool
+  calls. Onboarded projects previously could answer read-only prompts (Q&A,
+  "what is this project", explanations) without loading `CLAUDE.md`,
+  `.paqad/framework-path.txt`, or `docs/instructions/{rules,stack,design-system}`
+  — the framework's rules and Decision Pause Contract silently never entered
+  context.
+
+  The Claude Code adapter now installs a `UserPromptSubmit` hook
+  (`runtime/hooks/agent-entry-prompt-gate.sh`) alongside the existing
+  `PreToolUse` hook. Both gates share sentinel-freshness logic via
+  `runtime/hooks/lib/agent-entry-sentinel.sh` so they cannot drift.
+
+  Soft mode (default): the hook prints a high-priority reminder on stdout; Claude
+  Code injects it into the model context before the turn is planned. Hard mode
+  (`PAQAD_AGENT_ENTRY_MODE=hard`): the hook exits non-zero and blocks the turn
+  until the sentinel is written.
+
+  Re-running onboarding / `paqad upgrade` refreshes the wiring in
+  `.claude/settings.json` idempotently. Resolves Eliyce/paqad-ai#74.
+
 ## 1.3.0
 
 ### Minor Changes