trace2rules 0.1.0__tar.gz → 0.1.2__tar.gz

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 (21) hide show
  1. {trace2rules-0.1.0 → trace2rules-0.1.2}/PKG-INFO +11 -5
  2. {trace2rules-0.1.0 → trace2rules-0.1.2}/README.md +10 -4
  3. {trace2rules-0.1.0 → trace2rules-0.1.2}/pyproject.toml +1 -1
  4. {trace2rules-0.1.0 → trace2rules-0.1.2}/pyproject.toml.orig +1 -1
  5. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/cli.py +196 -34
  6. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/detectors.py +2 -0
  7. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/ingest.py +45 -30
  8. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/labels.py +3 -2
  9. trace2rules-0.1.2/src/trace2rules/redact.py +67 -0
  10. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/sources/agent_trajectory.py +152 -11
  11. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/synthesize.py +160 -51
  12. {trace2rules-0.1.0 → trace2rules-0.1.2}/LICENSE +0 -0
  13. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/__init__.py +0 -0
  14. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/cluster.py +0 -0
  15. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/db.py +0 -0
  16. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/models.py +0 -0
  17. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/patterns.py +0 -0
  18. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/sources/__init__.py +0 -0
  19. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/sources/common.py +0 -0
  20. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/sources/opencode.py +0 -0
  21. {trace2rules-0.1.0 → trace2rules-0.1.2}/src/trace2rules/sources/swe_chat.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: trace2rules
3
- Version: 0.1.0
3
+ Version: 0.1.2
4
4
  Summary: Mine coding-agent trajectories into evidence-backed AGENTS.md rules
5
5
  Keywords: agents,coding-agents,agents.md,trajectories,duckdb
6
6
  Author: Rakha Kanz Kautsar
@@ -76,7 +76,7 @@ Rules are capped at 5 rules, 60 lines, and 2,500 characters per repo.
76
76
 
77
77
  ## Synthesis
78
78
 
79
- Synthesis is LLM-first. `trace2rules synthesize` sends the gated findings for each repo to an OpenAI-compatible chat model and asks for structured rules.
79
+ Synthesis is LLM-first. `trace2rules synthesize` sends redacted gated findings for each repo to an OpenAI-compatible chat model and asks for structured rules. Common API keys, bearer tokens, URL credentials, query tokens, password assignments, and private keys are redacted before external calls; redaction is best-effort, not a guarantee.
80
80
 
81
81
  Provider selection:
82
82
 
@@ -119,7 +119,7 @@ uv sync
119
119
 
120
120
  ## Quick start: your own sessions
121
121
 
122
- Analyze the sessions already on your machine. This example uses OMP; point `--source` at Claude Code, Codex, or Pi instead if that is what you use.
122
+ Analyze the sessions already on your machine.
123
123
 
124
124
  ```bash
125
125
  uv tool install trace2rules
@@ -128,6 +128,10 @@ trace2rules mine
128
128
  trace2rules synthesize
129
129
  ```
130
130
 
131
+ Or hand the workflow to your coding agent:
132
+
133
+ > Read the `trace2rules` README, install it with `uv tool install trace2rules` if it is missing, then ingest my local agent sessions. Use `trace2rules report` to surface recurring friction with evidence and synthesize repo-specific `AGENTS.md` rules from that JSON.
134
+
131
135
  `--source` is repeatable. `--source ~` scans all standard local stores: `~/.claude/projects`, `~/.codex/sessions`, `~/.pi/agent/sessions`, and `~/.omp/agent/sessions`. The example imports 50 sessions per source; use `--limit 0` for all.
132
136
 
133
137
  LLM synthesis is the default. Add `--offline` if you do not have an API key.
@@ -200,6 +204,7 @@ Use `--repo owner/name` to override the repository label for local sessions.
200
204
  | `ingest` | Load sessions and tool events into DuckDB |
201
205
  | `mine` | Show friction clusters, hotspots, recovery loops, and context blasts |
202
206
  | `report` | Emit all findings and patterns as evidence-rich JSON for an agent |
207
+ | `doctor` | Check install, Node, credentials, source stores, and database schema |
203
208
  | `label` | Optional: classify targets with a decision model (`intent`, `object_type`, confidence) |
204
209
  | `patterns` | Mine structural multi-step friction; labels refine when available |
205
210
  | `exact` | Show exact recurring sequences with evidence |
@@ -214,9 +219,10 @@ trace2rules ingest --format swe-rebench # opt into the public dataset
214
219
  trace2rules ingest --source PATH --limit 200
215
220
  trace2rules synthesize
216
221
  trace2rules synthesize --offline
222
+ trace2rules synthesize --max-rules 10
223
+ trace2rules synthesize --out-dir ./proposals
217
224
  trace2rules synthesize --repo owner/name --out AGENTS.proposed.md
218
225
  trace2rules synthesize --repo owner/name --out -
219
- trace2rules synthesize --out-dir ./proposals
220
226
  ```
221
227
 
222
228
  `--label` auto-detects an authenticated `wrangler` first, then `GEMINI_API_KEY`/`GOOGLE_API_KEY`, then `OPENAI_API_KEY`. It is opt-in because a full local run can have thousands of unique targets.
@@ -269,7 +275,7 @@ trace2rules patterns --min-confidence 0.4
269
275
 
270
276
  Cloudflare models authenticate through `wrangler auth token --json` and `wrangler whoami --json`; the first account is used. Gemini and OpenAI use their standard keys. Labeling defaults to 270 requests/minute with 6 workers and retries 429/5xx responses.
271
277
 
272
- `label` sends the tool name, action type, normalized target, and a raw command example to the selected provider. Do not run it on private sessions without review.
278
+ `label` sends the tool name, action type, redacted normalized target, and a redacted raw command example to the selected provider. The local DuckDB and `report` output remain unredacted. Do not run external labeling or synthesis on private sessions without reviewing the redaction and provider policy.
273
279
 
274
280
  ## Configuration
275
281
 
@@ -47,7 +47,7 @@ Rules are capped at 5 rules, 60 lines, and 2,500 characters per repo.
47
47
 
48
48
  ## Synthesis
49
49
 
50
- Synthesis is LLM-first. `trace2rules synthesize` sends the gated findings for each repo to an OpenAI-compatible chat model and asks for structured rules.
50
+ Synthesis is LLM-first. `trace2rules synthesize` sends redacted gated findings for each repo to an OpenAI-compatible chat model and asks for structured rules. Common API keys, bearer tokens, URL credentials, query tokens, password assignments, and private keys are redacted before external calls; redaction is best-effort, not a guarantee.
51
51
 
52
52
  Provider selection:
53
53
 
@@ -90,7 +90,7 @@ uv sync
90
90
 
91
91
  ## Quick start: your own sessions
92
92
 
93
- Analyze the sessions already on your machine. This example uses OMP; point `--source` at Claude Code, Codex, or Pi instead if that is what you use.
93
+ Analyze the sessions already on your machine.
94
94
 
95
95
  ```bash
96
96
  uv tool install trace2rules
@@ -99,6 +99,10 @@ trace2rules mine
99
99
  trace2rules synthesize
100
100
  ```
101
101
 
102
+ Or hand the workflow to your coding agent:
103
+
104
+ > Read the `trace2rules` README, install it with `uv tool install trace2rules` if it is missing, then ingest my local agent sessions. Use `trace2rules report` to surface recurring friction with evidence and synthesize repo-specific `AGENTS.md` rules from that JSON.
105
+
102
106
  `--source` is repeatable. `--source ~` scans all standard local stores: `~/.claude/projects`, `~/.codex/sessions`, `~/.pi/agent/sessions`, and `~/.omp/agent/sessions`. The example imports 50 sessions per source; use `--limit 0` for all.
103
107
 
104
108
  LLM synthesis is the default. Add `--offline` if you do not have an API key.
@@ -171,6 +175,7 @@ Use `--repo owner/name` to override the repository label for local sessions.
171
175
  | `ingest` | Load sessions and tool events into DuckDB |
172
176
  | `mine` | Show friction clusters, hotspots, recovery loops, and context blasts |
173
177
  | `report` | Emit all findings and patterns as evidence-rich JSON for an agent |
178
+ | `doctor` | Check install, Node, credentials, source stores, and database schema |
174
179
  | `label` | Optional: classify targets with a decision model (`intent`, `object_type`, confidence) |
175
180
  | `patterns` | Mine structural multi-step friction; labels refine when available |
176
181
  | `exact` | Show exact recurring sequences with evidence |
@@ -185,9 +190,10 @@ trace2rules ingest --format swe-rebench # opt into the public dataset
185
190
  trace2rules ingest --source PATH --limit 200
186
191
  trace2rules synthesize
187
192
  trace2rules synthesize --offline
193
+ trace2rules synthesize --max-rules 10
194
+ trace2rules synthesize --out-dir ./proposals
188
195
  trace2rules synthesize --repo owner/name --out AGENTS.proposed.md
189
196
  trace2rules synthesize --repo owner/name --out -
190
- trace2rules synthesize --out-dir ./proposals
191
197
  ```
192
198
 
193
199
  `--label` auto-detects an authenticated `wrangler` first, then `GEMINI_API_KEY`/`GOOGLE_API_KEY`, then `OPENAI_API_KEY`. It is opt-in because a full local run can have thousands of unique targets.
@@ -240,7 +246,7 @@ trace2rules patterns --min-confidence 0.4
240
246
 
241
247
  Cloudflare models authenticate through `wrangler auth token --json` and `wrangler whoami --json`; the first account is used. Gemini and OpenAI use their standard keys. Labeling defaults to 270 requests/minute with 6 workers and retries 429/5xx responses.
242
248
 
243
- `label` sends the tool name, action type, normalized target, and a raw command example to the selected provider. Do not run it on private sessions without review.
249
+ `label` sends the tool name, action type, redacted normalized target, and a redacted raw command example to the selected provider. The local DuckDB and `report` output remain unredacted. Do not run external labeling or synthesis on private sessions without reviewing the redaction and provider policy.
244
250
 
245
251
  ## Configuration
246
252
 
@@ -1,6 +1,6 @@
1
1
  [project]
2
2
  name = "trace2rules"
3
- version = "0.1.0"
3
+ version = "0.1.2"
4
4
  description = "Mine coding-agent trajectories into evidence-backed AGENTS.md rules"
5
5
  readme = "README.md"
6
6
  license = "MIT"
@@ -1,6 +1,6 @@
1
1
  [project]
2
2
  name = "trace2rules"
3
- version = "0.1.0"
3
+ version = "0.1.2"
4
4
  description = "Mine coding-agent trajectories into evidence-backed AGENTS.md rules"
5
5
  readme = "README.md"
6
6
  license = "MIT"
@@ -2,8 +2,13 @@ from __future__ import annotations
2
2
 
3
3
  import json
4
4
  import os
5
+ import platform
6
+ import shutil
7
+ import subprocess
8
+ import sys
5
9
  from dataclasses import asdict
6
10
  from datetime import datetime, timezone
11
+ from importlib.metadata import PackageNotFoundError, version
7
12
  from pathlib import Path
8
13
  from typing import Annotated
9
14
 
@@ -12,14 +17,13 @@ import typer
12
17
  from dotenv import load_dotenv
13
18
  from rich.console import Console
14
19
  from rich.markup import escape
20
+ from rich.progress import BarColumn, Progress, SpinnerColumn, TextColumn, TimeElapsedColumn
15
21
  from rich.table import Table
16
22
 
17
- from .cluster import exact_sequences, mine as mine_friction
18
23
  from .db import DB_PATH, connect
19
24
  from .ingest import DEFAULT_REPOS, SOURCE, ingest as ingest_traces
20
25
  from .labels import label_targets
21
26
  from .patterns import mine_sequences
22
- from .synthesize import EXACT_LENGTHS, synthesize as synthesize_rules
23
27
 
24
28
  load_dotenv(Path.cwd() / ".env")
25
29
 
@@ -64,59 +68,69 @@ def _ingest_command(
64
68
  ) -> None:
65
69
  fmt = format.strip().lower()
66
70
  sources = source or []
71
+ failed_sources: list[tuple[str, Exception]] = []
67
72
  if repos and fmt != "swe-rebench":
68
73
  raise typer.BadParameter("--repos is only valid with --format swe-rebench.")
69
74
  if fmt == "swe-rebench":
70
75
  prefixes = repos or DEFAULT_REPOS
71
76
  selected_sources = sources or [SOURCE]
72
77
  stats = []
73
- try:
74
- for selected_source in selected_sources:
75
- if selected_source.startswith(("hf://", "http://", "https://")):
76
- console.print("[dim]Scanning remote parquet; first run downloads metadata/data...[/dim]")
78
+ for selected_source in selected_sources:
79
+ if selected_source.startswith(("hf://", "http://", "https://")):
80
+ console.print("[dim]Scanning remote parquet; first run downloads metadata/data...[/dim]")
81
+ try:
77
82
  stats.extend(
78
83
  ingest_traces(
79
84
  prefixes, limit=limit, db_path=db, source=selected_source
80
85
  )
81
86
  )
82
- except (duckdb.Error, OSError, RuntimeError) as error:
83
- console.print(f"[red]{escape(str(error))}[/red]")
84
- raise typer.Exit(1) from None
87
+ except (duckdb.Error, OSError, RuntimeError) as error:
88
+ failed_sources.append((selected_source, error))
85
89
  elif fmt == "swe-chat":
86
90
  from .sources.swe_chat import ingest_swe_chat
87
91
 
88
92
  selected_sources = sources or ["data/swe-chat"]
89
- try:
90
- stats = [
91
- ingest_swe_chat(
92
- selected_source,
93
- db,
94
- limit=limit or None,
95
- repo=repo,
93
+ stats = []
94
+ for selected_source in selected_sources:
95
+ try:
96
+ stats.append(
97
+ ingest_swe_chat(
98
+ selected_source,
99
+ db,
100
+ limit=limit or None,
101
+ repo=repo,
102
+ )
96
103
  )
97
- for selected_source in selected_sources
98
- ]
99
- except (FileNotFoundError, ValueError, RuntimeError) as error:
100
- console.print(f"[red]{escape(str(error))}[/red]")
101
- raise typer.Exit(1) from None
104
+ except (FileNotFoundError, ValueError, RuntimeError) as error:
105
+ failed_sources.append((selected_source, error))
102
106
  elif fmt in {"agent-trajectory", "agent"}:
103
107
  from .sources.agent_trajectory import ingest_agent_sessions
104
108
 
105
109
  selected_sources = sources or [str(Path.home())]
106
110
  stats = []
107
- try:
111
+ progress = Progress(
112
+ SpinnerColumn(),
113
+ TextColumn("[progress.description]{task.description}"),
114
+ BarColumn(),
115
+ TextColumn("{task.completed}/{task.total}"),
116
+ TimeElapsedColumn(),
117
+ console=Console(stderr=True),
118
+ transient=False,
119
+ )
120
+ with progress:
108
121
  for selected_source in selected_sources:
109
- stats.extend(
110
- ingest_agent_sessions(
111
- selected_source,
112
- db,
113
- limit=limit or None,
114
- repo=repo,
122
+ try:
123
+ stats.extend(
124
+ ingest_agent_sessions(
125
+ selected_source,
126
+ db,
127
+ limit=limit or None,
128
+ repo=repo,
129
+ progress=progress,
130
+ )
115
131
  )
116
- )
117
- except (FileNotFoundError, ValueError, RuntimeError) as error:
118
- console.print(f"[red]{escape(str(error))}[/red]")
119
- raise typer.Exit(1) from None
132
+ except (FileNotFoundError, ValueError, RuntimeError) as error:
133
+ failed_sources.append((selected_source, error))
120
134
  else:
121
135
  raise typer.BadParameter(
122
136
  "Unknown format. Use swe-rebench, swe-chat, or agent-trajectory."
@@ -136,6 +150,12 @@ def _ingest_command(
136
150
  for stat in stats:
137
151
  if stat.matched == 0:
138
152
  console.print(f"[yellow]No sessions found for {stat.prefix!r}.[/yellow]")
153
+ for failed_source, error in failed_sources:
154
+ console.print(
155
+ f"[red]Failed {escape(str(failed_source))}: {escape(str(error))}[/red]"
156
+ )
157
+ if failed_sources and not stats:
158
+ raise typer.Exit(1)
139
159
  if label:
140
160
  from .labels import detect_model
141
161
 
@@ -281,6 +301,7 @@ def _mine_command(
281
301
  console.print("[red]No sessions in the database. Run `trace2rules ingest` first.[/red]")
282
302
  raise typer.Exit(1)
283
303
  from .detectors import session_summary
304
+ from .cluster import mine as mine_friction
284
305
 
285
306
  _print_summary(session_summary(con))
286
307
  mined = mine_friction(con, top=top)
@@ -318,6 +339,9 @@ def _report_command(
318
339
  "SELECT repo, count(*) FROM sessions GROUP BY repo ORDER BY repo"
319
340
  ).fetchall()
320
341
 
342
+ from .cluster import exact_sequences, mine as mine_friction
343
+ from .synthesize import EXACT_LENGTHS
344
+
321
345
  mined = mine_friction(con, top=top)
322
346
  findings = [
323
347
  finding
@@ -368,6 +392,127 @@ def _report_command(
368
392
  app.command("report")(_report_command)
369
393
 
370
394
 
395
+ def _doctor_command(
396
+ db: Annotated[Path, typer.Option(help="DuckDB database path.")] = DB_PATH,
397
+ ) -> None:
398
+ checks: list[tuple[str, str, str]] = []
399
+
400
+ py_ok = sys.version_info >= (3, 12)
401
+ checks.append(("Python", "ok" if py_ok else "fail", platform.python_version()))
402
+
403
+ try:
404
+ package_version = version("trace2rules")
405
+ checks.append(("trace2rules", "ok", package_version))
406
+ except PackageNotFoundError:
407
+ checks.append(("trace2rules", "warn", "not installed as a package"))
408
+
409
+ node = shutil.which("node")
410
+ if not node:
411
+ checks.append(("Node.js", "warn", "not found; non-OpenCode local sources need it"))
412
+ else:
413
+ try:
414
+ node_version = subprocess.run(
415
+ [node, "--version"],
416
+ capture_output=True,
417
+ text=True,
418
+ timeout=5,
419
+ check=False,
420
+ ).stdout.strip()
421
+ except OSError:
422
+ node_version = "unavailable"
423
+ checks.append(("Node.js", "ok", node_version or node))
424
+
425
+ try:
426
+ from .labels import _wrangler_ready, detect_model
427
+
428
+ wrangler_ready = _wrangler_ready()
429
+ checks.append(
430
+ (
431
+ "wrangler",
432
+ "ok" if wrangler_ready else "warn",
433
+ "authenticated" if wrangler_ready else "not authenticated",
434
+ )
435
+ )
436
+ decision_model = detect_model()
437
+ checks.append(
438
+ (
439
+ "decision backend",
440
+ "ok" if decision_model else "warn",
441
+ decision_model or "none detected",
442
+ )
443
+ )
444
+ except Exception as error: # noqa: BLE001 - doctor must not fail on probes
445
+ checks.append(("decision backend", "warn", str(error)))
446
+
447
+ for key in ("TRACE2RULES_LLM_API_KEY", "GEMINI_API_KEY", "OPENAI_API_KEY"):
448
+ checks.append(
449
+ (key, "ok" if os.getenv(key) else "warn", "set" if os.getenv(key) else "unset")
450
+ )
451
+
452
+ from .sources.agent_trajectory import _source_roots
453
+
454
+ try:
455
+ roots = _source_roots(Path.home())
456
+ detail = ", ".join(f"{name}:{path}" for name, path in roots) or "none found"
457
+ checks.append(
458
+ ("local sources", "ok" if roots else "warn", detail)
459
+ )
460
+ except OSError as error:
461
+ checks.append(("local sources", "warn", str(error)))
462
+
463
+ parent = Path(db).parent
464
+ try:
465
+ parent.mkdir(parents=True, exist_ok=True)
466
+ writable = os.access(parent, os.W_OK)
467
+ except OSError:
468
+ writable = False
469
+ checks.append(
470
+ ("database directory", "ok" if writable else "fail", str(parent))
471
+ )
472
+ if Path(db).exists():
473
+ try:
474
+ con = duckdb.connect(str(db), read_only=True)
475
+ try:
476
+ tables = {
477
+ row[0]
478
+ for row in con.execute(
479
+ "SELECT table_name FROM information_schema.tables"
480
+ ).fetchall()
481
+ }
482
+ columns = {
483
+ row[1]
484
+ for row in con.execute("PRAGMA table_info('tool_events')").fetchall()
485
+ }
486
+ if "tool_events" not in tables:
487
+ status, detail = "fail", "missing tool_events"
488
+ elif "segment_idx" not in columns:
489
+ status, detail = "warn", "schema needs re-ingest/migration"
490
+ else:
491
+ status, detail = "ok", "schema ready"
492
+ checks.append(("database", status, detail))
493
+ finally:
494
+ con.close()
495
+ except Exception as error: # noqa: BLE001 - report lock/permission issues
496
+ checks.append(("database", "warn", str(error)))
497
+ else:
498
+ checks.append(("database", "warn", f"not created yet: {db}"))
499
+
500
+ table = Table(title="trace2rules doctor", header_style="bold")
501
+ table.add_column("Check")
502
+ table.add_column("Status")
503
+ table.add_column("Detail", overflow="fold")
504
+ for check, status, detail in checks:
505
+ style = {"ok": "green", "warn": "yellow", "fail": "red"}.get(status, "")
506
+ table.add_row(check, f"[{style}]{status}[/{style}]", escape(detail))
507
+ console.print(table)
508
+
509
+ if not py_ok or not writable:
510
+ raise typer.Exit(1)
511
+
512
+
513
+ app.command("doctor")(_doctor_command)
514
+
515
+
371
516
  def _label_command(
372
517
  limit: Annotated[int, typer.Option(help="Max pending targets; 0 = all.")] = 0,
373
518
  rpm: Annotated[int, typer.Option(help="Requests per minute cap.")] = 270,
@@ -456,6 +601,8 @@ def _exact_command(
456
601
  ) -> None:
457
602
  con = connect(db)
458
603
  try:
604
+ from .cluster import exact_sequences
605
+
459
606
  findings = exact_sequences(
460
607
  con,
461
608
  length=length,
@@ -500,15 +647,27 @@ app.command("exact")(_exact_command)
500
647
 
501
648
  def _synthesize_command(
502
649
  out: Annotated[Path | None, typer.Option("--out", "-o", help="Output path for a single --repo; use '-' for stdout.")] = None,
503
- out_dir: Annotated[Path, typer.Option("--out-dir", help="Directory for per-repo AGENTS files.")] = Path("."),
650
+ out_dir: Annotated[
651
+ Path | None,
652
+ typer.Option("--out-dir", help="Directory for per-repo AGENTS files."),
653
+ ] = None,
504
654
  repo: Annotated[str | None, typer.Option("--repo", help="Synthesize only this repo.")] = None,
505
655
  top: Annotated[int, typer.Option(help="Findings to pass to synthesis.")] = 25,
656
+ max_rules: Annotated[
657
+ int,
658
+ typer.Option("--max-rules", min=1, max=20, help="Maximum rules per repo."),
659
+ ] = 5,
506
660
  offline: Annotated[bool, typer.Option("--offline", help="Skip the LLM and use deterministic rules.")] = False,
507
661
  db: Annotated[Path, typer.Option(help="DuckDB database path.")] = DB_PATH,
508
662
  ) -> None:
509
663
  if out is not None and repo is None:
510
664
  raise typer.BadParameter("--out requires --repo; use --out-dir for multiple repos.")
511
- cmd_console = Console(stderr=True) if out is not None and str(out) == "-" else console
665
+ stdout_mode = out is None and out_dir is None
666
+ cmd_console = (
667
+ Console(stderr=True)
668
+ if stdout_mode or (out is not None and str(out) == "-")
669
+ else console
670
+ )
512
671
  con = connect(db)
513
672
  try:
514
673
  session_count = con.execute("SELECT count(*) FROM sessions").fetchone()[0]
@@ -516,12 +675,15 @@ def _synthesize_command(
516
675
  cmd_console.print("[red]No sessions in the database. Run `trace2rules ingest` first.[/red]")
517
676
  raise typer.Exit(1)
518
677
  try:
678
+ from .synthesize import synthesize as synthesize_rules
679
+
519
680
  results = synthesize_rules(
520
681
  con,
521
682
  out_path=out,
522
683
  out_dir=out_dir,
523
684
  repo=repo,
524
685
  top=top,
686
+ max_rules=max_rules,
525
687
  offline=offline,
526
688
  console=cmd_console,
527
689
  )
@@ -27,6 +27,7 @@ class Finding:
27
27
  intent_confidence: float = 0.0
28
28
  object_confidence: float = 0.0
29
29
  event_keys: tuple[tuple[str, str, str], ...] = field(default_factory=tuple)
30
+ edit_sessions: int = 0
30
31
 
31
32
  @property
32
33
  def evidence(self) -> str:
@@ -102,6 +103,7 @@ def hotspots(con: duckdb.DuckDBPyConnection, limit: int | None = None) -> list[F
102
103
  sessions=sessions,
103
104
  tokens=int(wasted or 0),
104
105
  detail=f"read-only in {read_only}/{repo_sessions} sessions ({read_only / repo_sessions:.0%})",
106
+ edit_sessions=sessions - read_only,
105
107
  )
106
108
  for repo, target, sessions, read_only, repo_sessions, wasted in con.execute(query).fetchall()
107
109
  ]
@@ -562,15 +562,22 @@ def write_sessions(
562
562
  sessions: Iterable[Session],
563
563
  ) -> IngestStats:
564
564
  """Write canonical sessions/events, replacing rows with the same ids."""
565
- session_list = list(sessions)
565
+ session_list = list({session.session_id: session for session in sessions}.values())
566
566
  con = connect(db_path)
567
- event_count = 0
567
+ con.execute("BEGIN TRANSACTION")
568
568
  try:
569
- for session in session_list:
570
- con.execute("DELETE FROM tool_events WHERE session_id = ?", [session.session_id])
571
- con.execute("DELETE FROM sessions WHERE session_id = ?", [session.session_id])
572
- total_turns = len(session.calls)
569
+ if session_list:
570
+ session_ids = [session.session_id for session in session_list]
571
+ placeholders = ",".join("?" for _ in session_ids)
572
+ con.execute(
573
+ f"DELETE FROM tool_events WHERE session_id IN ({placeholders})",
574
+ session_ids,
575
+ )
573
576
  con.execute(
577
+ f"DELETE FROM sessions WHERE session_id IN ({placeholders})",
578
+ session_ids,
579
+ )
580
+ con.executemany(
574
581
  """
575
582
  INSERT INTO sessions (
576
583
  session_id, repo, instance_id, resolved, total_turns,
@@ -578,17 +585,22 @@ def write_sessions(
578
585
  ) VALUES (?, ?, ?, ?, ?, ?, ?, ?)
579
586
  """,
580
587
  [
581
- session.session_id,
582
- session.repo or "unknown",
583
- session.instance_id or session.session_id,
584
- session.resolved,
585
- total_turns,
586
- session.source,
587
- session.agent,
588
- session.success_score,
588
+ (
589
+ session.session_id,
590
+ session.repo or "unknown",
591
+ session.instance_id or session.session_id,
592
+ session.resolved,
593
+ len(session.calls),
594
+ session.source,
595
+ session.agent,
596
+ session.success_score,
597
+ )
598
+ for session in session_list
589
599
  ],
590
600
  )
591
- events = []
601
+
602
+ events = []
603
+ for session in session_list:
592
604
  for step_idx, call in enumerate(session.calls):
593
605
  if call.action_type:
594
606
  action = call.action_type
@@ -643,29 +655,32 @@ def write_sessions(
643
655
  observation[:400],
644
656
  )
645
657
  )
646
- if events:
647
- con.executemany(
648
- """
649
- INSERT INTO tool_events (
650
- event_id, session_id, repo, step_idx, segment_idx,
651
- tool_name, action_type,
652
- verb, object_id, target, raw_target, outcome, exit_code,
653
- error_signature, is_error, approx_out_tokens, obs_excerpt
654
- ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
655
- """,
656
- events,
657
- )
658
- event_count += len(events)
658
+
659
+ if events:
660
+ con.executemany(
661
+ """
662
+ INSERT INTO tool_events (
663
+ event_id, session_id, repo, step_idx, segment_idx,
664
+ tool_name, action_type,
665
+ verb, object_id, target, raw_target, outcome, exit_code,
666
+ error_signature, is_error, approx_out_tokens, obs_excerpt
667
+ ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
668
+ """,
669
+ events,
670
+ )
671
+ con.execute("COMMIT")
659
672
  return IngestStats(
660
673
  prefix="canonical",
661
674
  matched=len(session_list),
662
675
  sessions=len(session_list),
663
- events=event_count,
676
+ events=len(events),
664
677
  )
678
+ except Exception:
679
+ con.execute("ROLLBACK")
680
+ raise
665
681
  finally:
666
682
  con.close()
667
683
 
668
-
669
684
  def _flush_sessions(con: duckdb.DuckDBPyConnection, sessions: list[dict]) -> None:
670
685
  if not sessions:
671
686
  return
@@ -14,6 +14,7 @@ from pathlib import Path
14
14
  import duckdb
15
15
 
16
16
  from .db import DB_PATH, connect
17
+ from .redact import redact_text
17
18
 
18
19
  DEFAULT_MODEL = "clef-flash"
19
20
  DEFAULT_RPM = 270
@@ -355,8 +356,8 @@ def _state(row: tuple[str, str, str, str]) -> dict:
355
356
  return {
356
357
  "tool": tool_name,
357
358
  "action_type": action_type,
358
- "normalized_target": target[:500],
359
- "raw_example": (raw_example or "").replace("\x00", " ")[:500],
359
+ "normalized_target": redact_text(target[:500]),
360
+ "raw_example": redact_text((raw_example or "").replace("\x00", " ")[:500]),
360
361
  }
361
362
 
362
363