@williambeto/ai-workflow 2.8.2 → 2.9.1

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 CHANGED
@@ -1,5 +1,63 @@
1
1
  ## [unreleased]
2
2
 
3
+ ## [2.9.1] - 2026-07-15
4
+
5
+ ### 🐛 Bug Fixes
6
+
7
+ - Accept persisted Playwright snapshots from the scoped delivery as valid UI
8
+ evidence, regardless of their test-directory location
9
+ - Recognize screenshots generated under `.evidence/visual/` and evaluate them
10
+ in the same `collect-evidence` execution that captured them
11
+ - Avoid repeating `NOT_DELIVERED` as a second blocking diagnostic when an
12
+ earlier validation or quality gate already explains the failure
13
+
14
+ ### 🧪 Testing
15
+
16
+ - Add regression coverage for Playwright snapshot discovery, generated visual
17
+ evidence ordering, missing required screenshots, and non-duplicated finalizer
18
+ diagnostics
19
+
20
+ ## [2.9.0] - 2026-07-15
21
+
22
+ ### 🚀 Features
23
+
24
+ - Add a Codex-native project installation with custom agents under
25
+ `.codex/agents/` and reusable skills, including commands, under
26
+ `.codex/skills/`
27
+ - Introduce runtime adapters and explicit `--runtime=opencode|codex` selection
28
+ while preserving OpenCode as the compatibility default
29
+ - Add Codex-aware doctor checks for trusted projects, native agents, skills,
30
+ policies, and CLI capabilities
31
+
32
+ ### 🐛 Bug Fixes
33
+
34
+ - Replace legacy Codex prompt and GitHub Copilot paths with documented native
35
+ Codex assets
36
+ - Preserve consumer-owned `AGENTS.md` files and install a Codex workflow
37
+ contract only when the root instructions are AIWK-managed or absent
38
+ - Block non-interactive Codex delivery when the runtime cannot prove custom
39
+ agent ownership instead of reporting unobserved Astra execution
40
+ - Keep runtime selection, delegation evidence, and bounded remediation aligned
41
+ across the OpenCode and Codex adapters
42
+
43
+ ### 🧪 Testing
44
+
45
+ - Add unit coverage for Codex agent conversion, command skills, runtime
46
+ selection, JSON event parsing, trust checks, and fail-closed execution
47
+ - Add a clean-consumer Codex init and doctor smoke test without generating
48
+ legacy prompt paths
49
+ - Validate the local release tarball in clean repositories with GPT-5.6 Sol
50
+ Medium and GPT-5.3 Codex Spark; qualify Sol for direct delivery and mark
51
+ Spark unsupported for workspace-write delivery after repeated code-only
52
+ responses without observed edits or validation
53
+
54
+ ### 📚 Documentation
55
+
56
+ - Reclassify Codex as experimental until an app-server E2E can prove native
57
+ delegation, workspace delivery, and observed validation
58
+ - Document project trust, native Codex asset locations, runtime selection, and
59
+ model-specific delivery limits
60
+
3
61
  ## [2.8.1] - 2026-07-14
4
62
 
5
63
  ### 🐛 Bug Fixes