selfware 0.6.1

Your personal AI workshop — software you own, software that lasts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211

#!/usr/bin/env python3
"""Evolve test harness — context-length matrix + level-1 pair suggestions.

Drives a running `selfware self-evolve` server. Two suites:

  A. Context-length matrix — 3 tasks x 5 context tiers (map -> full_extended),
     each a real config of orientation / comment-stripping / file budget, so the
     prompt the model receives grows monotonically. Records the actual prompt
     token length, latency, and whether the grounded review came back valid.

  B. Pair suggestions — the top cross-cluster level-1 pairs from the graph, each
     asked (via /api/evolve/pairs/suggest) for combined-evolution suggestions;
     validates the model returned a usable JSON suggestion list.

Both suites call the configured model (point the server at Kimi K3 via OpenRouter
first). `--check` runs ONLY the local endpoints (no model call) to validate the
wiring — pairs list, pair dry-run context, orientation preview, context sizes.

Usage:
  # validate wiring locally (no tokens spent):
  python3 scripts/kimi_evolve_matrix.py --check

  # full run against whatever model the server is configured with (Kimi K3):
  python3 scripts/kimi_evolve_matrix.py

  # narrow it:
  python3 scripts/kimi_evolve_matrix.py --tiers map,lite --pairs 2
"""
import argparse
import json
import sys
import time
import urllib.error
import urllib.request

# 3 tasks x 5 tiers. Each tier is a real context configuration; the label mirrors
# the composer tiers and the prompt-token column shows the length actually sent.
TASKS = [
    {"kind": "understand", "target": "agent::execution",
     "question": "What does this component do and how does it drive a task? Cite the source."},
    {"kind": "refactor", "target": "session::cache",
     "question": "How could this component be refactored for clarity without changing behavior?"},
    {"kind": "extend", "target": "tools::shell",
     "question": "How would I safely add a new capability here? Identify the seams."},
]

TIERS = {
    #  label            orient include_map compact max_files
    "map":           dict(orient=True,  include_map=True,  compact=True,  max_files=2),
    "lite":          dict(orient=True,  include_map=False, compact=True,  max_files=4),
    "compact":       dict(orient=False, include_map=False, compact=True,  max_files=8),
    "full":          dict(orient=False, include_map=False, compact=False, max_files=8),
    "full_extended": dict(orient=False, include_map=False, compact=False, max_files=20),
}
TIER_ORDER = ["map", "lite", "compact", "full", "full_extended"]


def req(base, path, method="GET", body=None, session=None, timeout=180):
    url = base.rstrip("/") + path
    data = json.dumps(body).encode() if body is not None else None
    headers = {"content-type": "application/json"}
    if session:
        headers["x-selfware-session"] = session
    r = urllib.request.Request(url, data=data, headers=headers, method=method)
    t0 = time.time()
    try:
        with urllib.request.urlopen(r, timeout=timeout) as resp:
            payload = json.loads(resp.read().decode())
            return payload, time.time() - t0, None
    except urllib.error.HTTPError as e:
        return None, time.time() - t0, f"HTTP {e.code}: {e.read().decode()[:200]}"
    except Exception as e:  # noqa: BLE001
        return None, time.time() - t0, str(e)


def workspace(base):
    payload, _, err = req(base, "/api/workspace")
    if err:
        sys.exit(f"cannot reach server at {base}: {err}")
    return payload


def suite_check(base):
    """Local-only wiring validation — no model calls, no tokens spent."""
    print("== CHECK (local endpoints, no model) ==")
    ok = True

    sizes, _, err = req(base, "/api/context/sizes")
    if err:
        ok = False; print("  context/sizes: FAIL", err)
    else:
        row = {s["mode"]: s["tokens"] for s in sizes["sizes"]}
        print("  context tiers:", ", ".join(f"{m}={row[m]:,}" for m in
              ["map", "lite", "compact", "full", "full_extended"] if m in row))

    for inc in ("false", "true"):
        o, _, err = req(base, f"/api/assistant/orientation?include_map={inc}")
        if err:
            ok = False; print(f"  orientation(map={inc}): FAIL", err)
        else:
            print(f"  orientation(map={inc}): {o['tokens']:,} tokens")

    pairs, _, err = req(base, "/api/evolve/pairs")
    if err:
        ok = False; print("  evolve/pairs: FAIL", err)
    else:
        print(f"  pairs: {pairs['pair_count']} ({pairs['cross_cluster']} cross-cluster)")
        top = pairs["pairs"][:3]
        for p in top:
            dry, _, err = req(base, "/api/evolve/pairs/suggest", "POST",
                              {"a": p["a"], "b": p["b"], "dry_run": True})
            if err:
                ok = False; print(f"    dry-run {p['a']}<->{p['b']}: FAIL", err)
            else:
                print(f"    dry-run {p['a']}<->{p['b']}: context {dry['context_tokens']:,} tok")
    print("  RESULT:", "OK" if ok else "FAIL")
    return ok


def suite_context_matrix(base, session, tiers):
    print("\n== A. CONTEXT-LENGTH MATRIX (3 tasks x tiers) ==")
    print(f"  {'task':<26} {'tier':<14} {'prompt_tok':>10} {'orient':>7} "
          f"{'lat_s':>6} {'valid':>5} {'claims':>6} {'recs':>4}")
    results = []
    for task in TASKS:
        label = f"{task['kind']}:{task['target']}"
        for tier in tiers:
            cfg = TIERS[tier]
            body = {**task, **cfg}
            payload, lat, err = req(base, "/api/assistant/task", "POST", body, session)
            if err:
                print(f"  {label:<26} {tier:<14} {'ERR':>10}  {err[:40]}")
                results.append({"task": label, "tier": tier, "error": err})
                continue
            review = payload.get("review", {})
            usage = review.get("usage", {})
            orient = payload.get("orientation", {})
            row = {
                "task": label, "tier": tier,
                "prompt_tokens": usage.get("prompt_tokens", 0),
                "orient_tokens": orient.get("tokens", 0),
                "latency_s": round(lat, 1),
                "valid": review.get("grounding_valid", False),
                "claims": len(review.get("claims", [])),
                "recs": len(review.get("recommendations", [])),
            }
            results.append(row)
            print(f"  {label:<26} {tier:<14} {row['prompt_tokens']:>10,} "
                  f"{row['orient_tokens']:>7,} {row['latency_s']:>6} "
                  f"{str(row['valid']):>5} {row['claims']:>6} {row['recs']:>4}")
    return results


def suite_pairs(base, session, n):
    print(f"\n== B. PAIR SUGGESTIONS (top {n} cross-cluster level-1 pairs) ==")
    pairs, _, err = req(base, "/api/evolve/pairs")
    if err:
        print("  cannot list pairs:", err); return []
    cross = [p for p in pairs["pairs"] if p["cross_cluster"]][:n]
    results = []
    for p in cross:
        payload, lat, err = req(base, "/api/evolve/pairs/suggest", "POST",
                                {"a": p["a"], "b": p["b"]}, session)
        tag = f"{p['a']}<->{p['b']}"
        if err:
            print(f"  {tag:<26} ERR  {err[:60]}")
            results.append({"pair": tag, "error": err}); continue
        sug = payload.get("suggestions") or {}
        items = sug.get("suggestions", []) if isinstance(sug, dict) else []
        valid = bool(items) and all("title" in s and s.get("steps") for s in items)
        usage = payload.get("usage", {})
        print(f"  {tag:<26} {len(items)} suggestions  valid={valid}  "
              f"ctx={payload.get('context_tokens', 0):,}tok  "
              f"prompt={usage.get('prompt_tokens', 0):,}tok  lat={lat:.1f}s")
        for s in items[:3]:
            print(f"      - [{s.get('kind', '?'):<13}] {s.get('title', '')} "
                  f"(effort {s.get('effort', '?')}, risk {s.get('risk', '?')})")
        results.append({"pair": tag, "count": len(items), "valid": valid})
    return results


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--base", default="http://127.0.0.1:8080")
    ap.add_argument("--check", action="store_true", help="local endpoints only, no model calls")
    ap.add_argument("--tiers", default=",".join(TIER_ORDER))
    ap.add_argument("--pairs", type=int, default=3)
    ap.add_argument("--skip-matrix", action="store_true")
    ap.add_argument("--skip-pairs", action="store_true")
    args = ap.parse_args()

    ws = workspace(args.base)
    print(f"server: {args.base}  model: {ws.get('model', '?')}  "
          f"context_length: {ws.get('context_length', '?')}")

    if args.check:
        sys.exit(0 if suite_check(args.base) else 1)

    session = ws.get("session_token")
    if not session:
        sys.exit("no session_token from /api/workspace")
    tiers = [t for t in args.tiers.split(",") if t in TIERS]

    if not args.skip_matrix:
        suite_context_matrix(args.base, session, tiers)
    if not args.skip_pairs:
        suite_pairs(args.base, session, args.pairs)


if __name__ == "__main__":
    main()