harness-evolver 1.1.0 → 1.2.0
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/package.json +1 -1
- package/skills/evolve/SKILL.md +37 -7
package/package.json
CHANGED
package/skills/evolve/SKILL.md
CHANGED
|
@@ -97,20 +97,50 @@ If score is 1.0 and iterations < 3, STOP the loop and strongly recommend the cri
|
|
|
97
97
|
> the eval is too easy, not that the harness is perfect. Run `/harness-evolver:critic`
|
|
98
98
|
> before continuing.
|
|
99
99
|
|
|
100
|
-
### 7.
|
|
100
|
+
### 7. Auto-trigger Architect (on stagnation or regression)
|
|
101
|
+
|
|
102
|
+
Check if the architect should be auto-spawned. This happens when:
|
|
103
|
+
- **Stagnation**: 3 consecutive iterations within 1% of each other
|
|
104
|
+
- **Regression**: score dropped below parent score (even once)
|
|
105
|
+
|
|
106
|
+
AND `.harness-evolver/architecture.json` does NOT already exist.
|
|
107
|
+
|
|
108
|
+
If triggered:
|
|
109
|
+
|
|
110
|
+
```bash
|
|
111
|
+
python3 $TOOLS/analyze_architecture.py \
|
|
112
|
+
--harness .harness-evolver/harnesses/{best_version}/harness.py \
|
|
113
|
+
--traces-dir .harness-evolver/harnesses/{best_version}/traces \
|
|
114
|
+
--summary .harness-evolver/summary.json \
|
|
115
|
+
-o .harness-evolver/architecture_signals.json
|
|
116
|
+
```
|
|
117
|
+
|
|
118
|
+
Then spawn the `harness-evolver-architect` agent:
|
|
119
|
+
|
|
120
|
+
> The evolution loop has stagnated/regressed after {iterations} iterations (best: {best_score}).
|
|
121
|
+
> Analyze the harness architecture and recommend a topology change.
|
|
122
|
+
> Raw signals at `.harness-evolver/architecture_signals.json`.
|
|
123
|
+
> Write `.harness-evolver/architecture.json` and `.harness-evolver/architecture.md`.
|
|
124
|
+
|
|
125
|
+
After the architect completes, report:
|
|
126
|
+
|
|
127
|
+
> Architect recommends: {current} → {recommended} ({confidence} confidence)
|
|
128
|
+
> Migration path: {N} steps. Continuing evolution with architecture guidance.
|
|
129
|
+
|
|
130
|
+
Then **continue the loop** — the proposer will read `architecture.json` in the next iteration.
|
|
131
|
+
|
|
132
|
+
If `architecture.json` already exists (architect already ran), skip — don't re-run.
|
|
133
|
+
|
|
134
|
+
### 8. Check Stop Conditions
|
|
101
135
|
|
|
102
|
-
- **Stagnation**: last 3 scores within 1% of each other → stop
|
|
103
136
|
- **Target**: `combined_score >= target_score` → stop
|
|
104
137
|
- **N reached**: done
|
|
138
|
+
- **Stagnation post-architect**: 3 more iterations without improvement AFTER architect ran → stop (architecture change didn't help)
|
|
105
139
|
|
|
106
140
|
## When Loop Ends — Final Report
|
|
107
141
|
|
|
108
142
|
- Best version and score
|
|
109
143
|
- Improvement over baseline (absolute and %)
|
|
110
144
|
- Total iterations run
|
|
145
|
+
- Whether architect was triggered and what it recommended
|
|
111
146
|
- Suggest: "The best harness is at `.harness-evolver/harnesses/{best}/harness.py`. Copy it to your project."
|
|
112
|
-
|
|
113
|
-
If the loop stopped due to stagnation AND `.harness-evolver/architecture.json` does NOT exist:
|
|
114
|
-
|
|
115
|
-
> The proposer may have hit an architectural ceiling. Run `/harness-evolver:architect`
|
|
116
|
-
> to analyze whether a different agent topology could help.
|