paqad-ai 1.90.0 → 1.91.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +25 -0
- package/dist/cli/index.d.ts +2 -0
- package/dist/cli/index.js +3164 -2252
- package/dist/cli/index.js.map +1 -1
- package/dist/feature-evidence/adoption.js.map +1 -1
- package/dist/index.d.ts +39 -5
- package/dist/index.js +2371 -1476
- package/dist/index.js.map +1 -1
- package/dist/kernel/gate.d.ts +4 -0
- package/dist/kernel/gate.js +4647 -3978
- package/dist/kernel/gate.js.map +1 -1
- package/dist/pipeline/dirty-baseline.d.ts +29 -0
- package/dist/pipeline/dirty-baseline.js +537 -0
- package/dist/pipeline/dirty-baseline.js.map +1 -0
- package/dist/pipeline/prompt-lane.js +105 -34
- package/dist/pipeline/prompt-lane.js.map +1 -1
- package/dist/rule-scripts/index.js +8 -3
- package/dist/rule-scripts/index.js.map +1 -1
- package/dist/runtime-paths-D7pDu5IJ.d.ts +16 -0
- package/dist/stage-evidence/live-writer.js +8 -3
- package/dist/stage-evidence/live-writer.js.map +1 -1
- package/dist/stage-evidence/marker-parse.js +8 -3
- package/dist/stage-evidence/marker-parse.js.map +1 -1
- package/dist/stage-evidence/narration.js +8 -3
- package/dist/stage-evidence/narration.js.map +1 -1
- package/dist/stage-isolation/subagent-completion.js +8 -3
- package/dist/stage-isolation/subagent-completion.js.map +1 -1
- package/package.json +1 -1
- package/runtime/hooks/agent-entry-prompt-gate.mjs +9 -0
- package/runtime/hooks/agent-entry-session-start.mjs +8 -0
- package/runtime/hooks/capability-gate.mjs +16 -1
- package/runtime/hooks/lib/agent-entry-directive.mjs +7 -3
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,30 @@
|
|
|
1
1
|
# paqad-ai
|
|
2
2
|
|
|
3
|
+
## 1.91.0
|
|
4
|
+
|
|
5
|
+
### Minor Changes
|
|
6
|
+
|
|
7
|
+
- 5c7b0fb: Visual evidence now actually runs from the Stop hook, and a frontend change hears about it at every stage (#579).
|
|
8
|
+
|
|
9
|
+
- **The gate fires again.** The Stop hook loads `dist/index.js`, one folder shallower than the CLI, and the package root was worked out with a fixed `../..`. From there it pointed one folder too high, no stack packs loaded, and every change read as "not frontend". paqad now walks up to its own `package.json`, so every bundle finds the same runtime folder. A new end-to-end test runs the built bundles and the real Stop-hook script to keep it that way.
|
|
10
|
+
- **A broken install says so.** If your profile names frameworks but paqad loads none of its packs, the visual-evidence gate reports an install fault (fail under strict, inconclusive under warn) and `visual-evidence run` / `plan` exit 1, instead of quietly passing the change.
|
|
11
|
+
- **New: `paqad-ai visual-evidence attach <png...> [--ac AC-3] [--label "..."]`.** When a scripted capture cannot run here, attach your own PNG screenshots. They are hashed and recorded as agent-attached (never shown as scripted captures), and a later `visual-evidence run` keeps them. `--ac` must name an acceptance criterion in the frozen spec, so a typo is refused instead of tying screenshots to nothing. A failed scripted step keeps the result partial whichever of `run` and `attach` came first, so attaching screenshots never hides a real gap.
|
|
12
|
+
- **Strict mode is stricter.** Under `visual_evidence_mode=strict`, a frontend change with nothing captured because there is no documented flow or capture script now fails. Attach screenshots, or record a waiver decision resolved with `waive`: the planning readiness decision when one was opened, otherwise one you create with `paqad-ai decision create` (the failing gate prints the exact command). A waived change reads skipped, never pass. Warn mode is unchanged and now prints a "⚪ visual evidence: skipped (reason)" line.
|
|
13
|
+
- **Asked for earlier.** `plan compile` opens one decision pause when the change is frontend and this machine cannot capture screenshots yet (plan steps can now list the `files` they touch). `spec freeze` asks for at least one `(proof: visual)` criterion on a frontend change. The first frontend edit carries a one-time reminder to the model.
|
|
14
|
+
- **On the record.** The late gates (bundle-completeness, visual-evidence, rules-loaded) are now written to the bundle's `evidence.jsonl` when the evidence ledger is on, skips included, with a new `skipped` verdict that never counts as a pass or a fail. `audit export` grades a skipped row informational (OCSF severity 1, CEF 1), and the evidence packet shows it with a grey dot instead of the amber "needs a look". A passing gate now names the `(proof: visual)` criteria it covers.
|
|
15
|
+
|
|
16
|
+
## 1.90.1
|
|
17
|
+
|
|
18
|
+
### Patch Changes
|
|
19
|
+
|
|
20
|
+
- c72b67b: fix(#576): make `paqad-ai join` fully provision a teammate machine and stop the first prompt of a session from falsely blocking at Stop
|
|
21
|
+
|
|
22
|
+
- Route every prompt (including the not-yet-loaded first prompt) so a read-only turn is classified non-feature and the Stop backstop no longer blames pre-existing dirty tracked files; also capture a session-start dirty-file baseline the completion backstop subtracts, and treat `.ai/` as host-agent config.
|
|
23
|
+
- `join` now performs the global framework install (`~/.paqad-ai/current` + stage agents), regenerates the per-machine detection/stack-snapshot/stack-drift artifacts, builds the code-knowledge index (also on `onboard`), runs delivery detection, seeds the quality baseline from HEAD, and rebuilds the RAG index without rewriting tracked files.
|
|
24
|
+
- The classifier routes interrogative project questions (how/where/what/why/explain/show me/example) to `project-question`.
|
|
25
|
+
- Git-hook installation skips a git-tracked hooks dir and reports the snippet instead; `join` closes four small correctness gaps (dead sentinel write, framework-version epoch seed, stale host-config refresh, tracked-aware ignore check); and the load directive steers the agent to its file-read tool to avoid the zsh `=` trap.
|
|
26
|
+
- `doctor`'s missing-artifact remediation now points at `paqad-ai join`.
|
|
27
|
+
|
|
3
28
|
## 1.90.0
|
|
4
29
|
|
|
5
30
|
### Minor Changes
|
package/dist/cli/index.d.ts
CHANGED
|
@@ -1,4 +1,6 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
|
+
export { a as getRuntimeRoot } from '../runtime-paths-D7pDu5IJ.js';
|
|
3
|
+
|
|
2
4
|
declare function getCliBanner(): string;
|
|
3
5
|
declare function runCli(argv?: string[]): Promise<void>;
|
|
4
6
|
declare function shouldRunFromCommandLine(importMetaUrl: string, argvEntry: string | undefined): boolean;
|