paqad-ai 1.0.6 → 1.2.0

package/CHANGELOG.md

@@ -1,5 +1,52 @@
 # paqad-ai
 
+## 1.2.0
+
+### Minor Changes
+
+- [#60](https://github.com/Eliyce/paqad-ai/pull/60) [`2e35581`](https://github.com/Eliyce/paqad-ai/commit/2e355817f14792fc51b7beada9001f96eaac4b13) Thanks [@HLasani](https://github.com/HLasani)! - **[#37](https://github.com/Eliyce/paqad-ai/issues/37) — Decision Pause Contract: ship the full resolution flow to every provider via one managed external doc**
+  - Onboarding now writes a canonical `.paqad/decision-pause-contract.md` (rule, categories sourced from `DECISION_CATEGORIES`, four-step resolution flow, per-adapter UI table, file-wait fallback) — re-runs are byte-identical when nothing changed.
+  - Every provider entry file (`CLAUDE.md`, `AGENTS.md`, `.cursor/rules/…`, `.windsurf/rules/…`, `GEMINI.md`, `.junie/AGENTS.md`, `ANTIGRAVITY.md`, `.github/copilot-instructions.md`, `.continue/…`, `.aider.conf.yml`) now renders a thin pointer + a one-sentence per-adapter UI note (Claude Code → `AskUserQuestion`, Aider → `/ask`, etc.) from the new `decision-pause-ui-shim`.
+  - `paqad refresh --providers` re-renders entry files for adapters whose config exists on disk and rewrites the managed doc; it intentionally does not silently onboard new providers.
+  - Health check now compares each entry against its adapter's expected rendering and warns when the managed doc is missing, with a remediation hint pointing at `paqad refresh --providers`.
+
+  **[#42](https://github.com/Eliyce/paqad-ai/issues/42) — `ticket_intake` + `delivery` bookend stages with priors-first decision elicitation**
+  - New stage order: `ticket_intake → planning → specification → development → review → checks → documentation_sync → delivery`. Both bookends are framework-owned and project-overridable via the existing `merge_mode: append` mechanism; the JSON schema rejects unknown keys with `additionalProperties: false`.
+  - New optional top-level `conventions:` block on the project profile covering `ticket / intake_decisions / branch / commit / pr`. `DEFAULT_CONVENTIONS` is the runtime source of truth; `resolveConventions` shallow-merges project overrides per section so every field is populated for downstream consumers.
+  - `mcp.servers[]` gains an optional `kind` discriminator (`jira | linear | github-issues | generic`); existing servers without `kind` keep validating.
+  - Four new `DECISION_CATEGORIES`: `intake.requirement`, `intake.confirm_auto_resolution`, `intake.write_back`, `delivery.open_pr`. All ride the existing pending → resolved lifecycle, audit log, and TTL machinery.
+  - `findIntakePriorMatch` is the dedicated priors-first entrypoint for intake; it computes the fingerprint and reuses the existing `DecisionStore.findReusableDecision` machinery, which already emits `decision-reused` audit events on every hit.
+  - New batched-confirm primitive (`BatchedConfirmRequest`, `applyBatchedAnswer`, `renderBatchedRow`) backs `intake.confirm_auto_resolution`. Single-packet flow remains the default for every other category.
+  - New `src/delivery/` modules: `templates.ts` renders branch / commit / pr_title / pr_body from the conventions block + inputs (slugify, conventional type mapping, `{placeholder}` substitution); `host.ts` detects the delivery host (GitHub today, GitLab / Bitbucket recognised); `runner.ts` sequences `git checkout -b → git commit → git push → gh pr create` through a dependency-injected `DeliveryShell` so unit tests don't touch a real remote. Every failure short-circuits with an actionable remediation hint — no silent local-only fallback.
+  - Default PR body template at `runtime/templates/pr-body.md.hbs`; projects can override with `.paqad/templates/pr-body.md`.
+  - Module docs: `docs/modules/decision-pause-contract/managed-doc-architecture.md` and `docs/modules/feature-development-workflow/ticket-intake-and-delivery.md`.
+
+## 1.1.0
+
+### Minor Changes
+
+- [#58](https://github.com/Eliyce/paqad-ai/pull/58) [`195ae7d`](https://github.com/Eliyce/paqad-ai/commit/195ae7dbca635305e8a3aaff9fb08f0af72e9cbc) Thanks [@HLasani](https://github.com/HLasani)! - Add a harness-enforced agent-entry gate so onboarded Claude Code projects can't silently bypass `CLAUDE.md` ([#34](https://github.com/Eliyce/paqad-ai/issues/34)). Onboarding now writes `.claude/settings.json` with two hooks:
+  - `PreToolUse` on `Edit|Write|NotebookEdit` runs `runtime/hooks/agent-entry-gate.sh`, which blocks the call with exit code 2 unless `.paqad/.agent-entry-loaded` exists and is newer than `CLAUDE.md`, `.paqad/framework-path.txt`, and everything under `docs/instructions/`.
+  - `SessionStart` runs `runtime/hooks/agent-entry-session-start.sh`, which deletes the sentinel so every new session starts ungated.
+
+  The CLAUDE.md template now instructs the agent to write the sentinel after loading the framework entry, and to re-load when the gate invalidates it. Existing settings.json keys and pre-existing hook entries are preserved on merge, and re-running onboarding is idempotent (no duplicate gate entries). Read-only tools (Read, Grep, Glob, status-only Bash) remain available pre-gate so the agent can satisfy it. AGENTS.md and other providers can re-use the same scripts in follow-ups via `PAQAD_ENTRY_FILE`.
+
+### Patch Changes
+
+- [#58](https://github.com/Eliyce/paqad-ai/pull/58) [`195ae7d`](https://github.com/Eliyce/paqad-ai/commit/195ae7dbca635305e8a3aaff9fb08f0af72e9cbc) Thanks [@HLasani](https://github.com/HLasani)! - Make onboarding re-runs idempotent ([#27](https://github.com/Eliyce/paqad-ai/issues/27)): when `.paqad/detection-report.json`, `.paqad/onboarding-manifest.json`, and `.paqad/framework-version.txt` would otherwise be byte-identical apart from their embedded timestamp, the writer now reuses the existing timestamp instead of stamping a fresh one. A no-op re-run produces zero diff; any real change still bumps the timestamp. Fixes the Windows-CI `is idempotent across repeated onboarding runs` failure (timing luck was hiding the same bug on macOS/Linux).
+
+- [#58](https://github.com/Eliyce/paqad-ai/pull/58) [`195ae7d`](https://github.com/Eliyce/paqad-ai/commit/195ae7dbca635305e8a3aaff9fb08f0af72e9cbc) Thanks [@HLasani](https://github.com/HLasani)! - `paqad-ai refresh` now self-heals when `docs/instructions/architecture/design-tokens.json` is missing ([#56](https://github.com/Eliyce/paqad-ai/issues/56)). The design-system step seeds default tokens via the existing idempotent `DesignTokenService.seed()` before generating docs and theme exports, so `--stack` and `--context` sub-refreshes no longer get aborted by an unhandled `ENOENT`. `DesignTokenService.load()` now translates a missing file into a typed `DesignTokensMissingError` with an actionable message; invalid (but present) files still surface the existing validation error.
+
+- [#58](https://github.com/Eliyce/paqad-ai/pull/58) [`195ae7d`](https://github.com/Eliyce/paqad-ai/commit/195ae7dbca635305e8a3aaff9fb08f0af72e9cbc) Thanks [@HLasani](https://github.com/HLasani)! - Stop emitting the hardcoded `Categories:` block in provider entry files ([#54](https://github.com/Eliyce/paqad-ai/issues/54)). `buildDecisionPauseContractSection()` now renders just the Decision Pause Contract paragraph — no `Categories:` heading, no fixed five-item bullet list. `extractDecisionPauseContractSection` still parses entry files that previously contained the block (back-compat), and the existing drift health check flags the legacy block so re-running onboarding / refresh strips it from already-onboarded projects.
+
+- [#58](https://github.com/Eliyce/paqad-ai/pull/58) [`195ae7d`](https://github.com/Eliyce/paqad-ai/commit/195ae7dbca635305e8a3aaff9fb08f0af72e9cbc) Thanks [@HLasani](https://github.com/HLasani)! - Cross-platform output consistency: normalize path strings to forward slashes at more production output boundaries — DocumentationWorkflow generated/skipped arrays and handover_path, onboarding orchestrator manifest writes and return values, `saveObligationIndex` return, registry-generator source_paths (both native-module and signal-extracted), and the project-question phase write target. Fix `slugifySpec` to split on both `/` and `\\` so spec-indexed compliance paths derive correctly on Windows. Rewrite `sanitizePersistedPath` to avoid relying on `process.cwd()` for relative inputs. Drops the Windows CI failure count from 15 → 11 test files ([#41](https://github.com/Eliyce/paqad-ai/issues/41)).
+
+## 1.0.7
+
+### Patch Changes
+
+- [#55](https://github.com/Eliyce/paqad-ai/pull/55) [`d47cf96`](https://github.com/Eliyce/paqad-ai/commit/d47cf964d0c661c8369e7f69c374df02ef84a8c9) Thanks [@HLasani](https://github.com/HLasani)! - Cross-platform output consistency: normalize path strings to forward slashes at more production output boundaries — DocumentationWorkflow generated/skipped arrays and handover_path, onboarding orchestrator manifest writes and return values, `saveObligationIndex` return, registry-generator source_paths (both native-module and signal-extracted), and the project-question phase write target. Fix `slugifySpec` to split on both `/` and `\\` so spec-indexed compliance paths derive correctly on Windows. Rewrite `sanitizePersistedPath` to avoid relying on `process.cwd()` for relative inputs. Drops the Windows CI failure count from 15 → 11 test files ([#41](https://github.com/Eliyce/paqad-ai/issues/41)).
+
 ## 1.0.6
 
 ### Patch Changes