compound-agent 2.5.0 → 2.5.2

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.
Files changed (2) hide show
  1. package/CHANGELOG.md +24 -0
  2. package/package.json +5 -5
package/CHANGELOG.md CHANGED
@@ -9,6 +9,30 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
9
9
 
10
10
  ## [Unreleased]
11
11
 
12
+ ## [2.5.2] - 2026-03-31
13
+
14
+ ### Added
15
+
16
+ - **Stale output watchdog (Layer 4)**: New background watchdog monitors trace file for output inactivity during Claude sessions. If no output is written for `SESSION_STALE_TIMEOUT` seconds (default: 1800s/30min), the session is killed and the loop proceeds. Prevents indefinite hangs when the Claude CLI completes its API work but fails to exit.
17
+
18
+ ### Fixed
19
+
20
+ - **Polish loop deadlock**: Architect prompt no longer uses `Parent: $META_EPIC` label, which caused the architect to wire blocking dependencies to a meta-epic that never closes. Replaced with context-only reference and explicit prohibition against `--parent` and `bd dep add` to the meta-epic.
21
+ - **Silent zero-work exit**: Infinity loop now exits with code 2 (distinct from success=0 and failure=1) when zero epics are completed and zero failed, signaling that all epics were blocked or skipped.
22
+ - **Inner loop exit code swallowed**: Polish loop's `run_inner_loop` used `|| true` which discarded the exit code. Now captures exit code properly and detects zero-work (exit 2) to surface blocked-epic deadlocks.
23
+ - **Inner loop `set -e` cascade**: `run_inner_loop` call in main polish loop now guarded with `||` handler, matching the pattern used for `run_polish_architect`, preventing a single failed cycle from aborting the entire polish script.
24
+
25
+ ### Removed
26
+
27
+ - **Improve loop references from architect skill**: Removed `ca improve` references from shipped architect SKILL.md, infinity-loop README, pre-flight docs, and deleted the `references/improve-loop/` directory. The improve loop remains available as a standalone CLI command but is not part of the architect workflow.
28
+
29
+ ## [2.5.1] - 2026-03-28
30
+
31
+ ### Added
32
+
33
+ - **Visual verification in polish audit**: Reviewers now auto-detect UI projects and take Playwright screenshots at 4 viewports (375px, 768px, 1024px, 1440px) as part of their audit, critiquing layout, spacing, contrast, hierarchy, and responsiveness alongside static code analysis. Graceful degradation when Playwright is unavailable or no UI detected (P3/INFO + [NEEDS_QA] fallback).
34
+ - **Visual verification in cook-it review**: Step 10b in the review skill takes Playwright screenshots when the Verification Contract requires `browser_evidence`, `design_craft_check`, or `responsive_check`. References QA Engineer detection logic for framework auto-detection.
35
+
12
36
  ## [2.5.0] - 2026-03-27
13
37
 
14
38
  ### Added
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "compound-agent",
3
- "version": "2.5.0",
3
+ "version": "2.5.2",
4
4
  "type": "module",
5
5
  "description": "Learning system for Claude Code — avoids repeating mistakes across sessions",
6
6
  "bin": {
@@ -50,10 +50,10 @@
50
50
  "knowledge-management"
51
51
  ],
52
52
  "optionalDependencies": {
53
- "@syottos/darwin-arm64": "2.5.0",
54
- "@syottos/darwin-x64": "2.5.0",
55
- "@syottos/linux-arm64": "2.5.0",
56
- "@syottos/linux-x64": "2.5.0"
53
+ "@syottos/darwin-arm64": "2.5.2",
54
+ "@syottos/darwin-x64": "2.5.2",
55
+ "@syottos/linux-arm64": "2.5.2",
56
+ "@syottos/linux-x64": "2.5.2"
57
57
  },
58
58
  "author": "Nathan Delacrétaz",
59
59
  "license": "MIT",