amazon-science/ammo / ai_cli_session/.codex/skills/ammo
ammo
UnverifiedProfile and optimize GPU kernels for vLLM inference on NVIDIA GPUs. Use when targeting a specific model, hardware, dtype, parallelism, and workload to improve latency.
eval/scripts/snapshot_changes.py
View on GitHub#!/usr/bin/env python3
# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: CC-BY-NC-4.0
"""Snapshot all code changes from an AMMO campaign for eval archiving.
Reads state.json to identify campaign worktrees, then captures:
- Full git diff (patch) for each worktree vs main
- Commit log for each worktree
- Worktree metadata (branch, merge status, changed files)
- Complete copy of the artifact directory (skipping huge binary files)
The output directory can be passed to archive_run.py via --changes-snapshot.
Usage:
python snapshot_changes.py \
--artifact-dir /path/to/kernel_opt_artifacts/auto_... \
--output /tmp/ammo_eval_changes_snapshot
"""
from __future__ import annotations
import argparse
import json
import shutil
import subprocess
import sys
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Dict, List, Optional
def _run_git(
args: List[str], cwd: Optional[str] = None, timeout: int = 30
) -> Optional[str]:
"""Run a git command and return stdout, or None on failure."""
try:
result = subprocess.run(
["git"] + args,
capture_output=True, text=True, timeout=timeout, cwd=cwd,
)
if result.returncode == 0:
return result.stdout.strip()
except (subprocess.TimeoutExpired, FileNotFoundError):
pass
return None
def _find_campaign_worktrees(
artifact_dir: Path, repo_root: Path
) -> List[Dict[str, str]]:
"""Find campaign worktrees from every round's parallel_tracks.tracks."""
state_path = artifact_dir / "state.json"
if not state_path.exists():
return []
state = json.loads(state_path.read_text(encoding="utf-8"))
rounds = state.get("campaign", {}).get("rounds", [])
worktrees = []
seen_paths = set()
for rnd in rounds:
if not isinstance(rnd, dict):
continue
tracks = rnd.get("parallel_tracks", {}).get("tracks", {})
if not isinstance(tracks, dict):
continue
for track_id, track_data in tracks.items():
if not isinstance(track_data, dict):
continue
wt_path = track_data.get("worktree") or track_data.get("worktree_path")
branch = track_data.get("branch") or track_data.get("worktree_branch")
if wt_path and wt_path not in seen_paths:
seen_paths.add(wt_path)
worktrees.append({
"path": wt_path,
"branch": branch or "unknown",
"track_id": track_id,
"status": track_data.get("status", "UNKNOWN"),
})
# Also check for additional worktrees under .codex/worktrees/ that match
# the same session but aren't in parallel_tracks (e.g. researcher agents)
session_ids = set()
for wt in worktrees:
branch = wt.get("branch", "")
if "session/" in branch:
parts = branch.split("/")
if len(parts) >= 2:
session_ids.add(parts[1])
if session_ids:
for wt_parent in [repo_root / ".codex" / "worktrees"]:
if not wt_parent.exists():
continue
for wt_dir in sorted(wt_parent.iterdir()):
if not wt_dir.is_dir():
continue
wt_str = str(wt_dir)
if wt_str in seen_paths:
continue
branch = _run_git(
["rev-parse", "--abbrev-ref", "HEAD"], cwd=wt_str
)
if not branch:
continue
for sid in session_ids:
if sid in branch:
seen_paths.add(wt_str)
worktrees.append({
"path": wt_str,
"branch": branch,
"track_id": None,
"status": "session-member",
})
break
return worktrees
def _capture_worktree_snapshot(
wt_info: Dict[str, str], main_branch: str = "main"
) -> Dict[str, Any]:
"""Capture full metadata + patch for a single worktree."""
wt_path = wt_info["path"]
if not Path(wt_path).exists():
return {
**wt_info,
"error": "worktree path does not exist",
"commit": None,
"patch": None,
}
commit = _run_git(["rev-parse", "HEAD"], cwd=wt_path) or "unknown"
short_hash = _run_git(["rev-parse", "--short", "HEAD"], cwd=wt_path) or "unknown"
# Commits since diverging from main
merge_base = _run_git(["merge-base", main_branch, "HEAD"], cwd=wt_path)
if merge_base:
log = _run_git(["log", "--oneline", f"{merge_base}..HEAD"], cwd=wt_path) or ""
diff = _run_git(
["diff", merge_base, "HEAD"], cwd=wt_path, timeout=60
) or ""
else:
log = _run_git(["log", "--oneline", "-10"], cwd=wt_path) or ""
diff = _run_git(
["diff", f"{main_branch}...HEAD"], cwd=wt_path, timeout=60
) or ""
# Uncommitted changes on top
uncommitted = _run_git(["diff", "HEAD"], cwd=wt_path, timeout=60) or ""
# Check merge status
is_merged = False
if merge_base:
check = _run_git(
["branch", "--contains", commit, main_branch], cwd=wt_path
)
is_merged = check is not None and main_branch in (check or "")
# Changed files list
if merge_base:
changed = _run_git(
["diff", "--name-only", merge_base, "HEAD"], cwd=wt_path
) or ""
else:
changed = _run_git(
["diff", "--name-only", f"{main_branch}...HEAD"], cwd=wt_path
) or ""
return {
"name": Path(wt_path).name,
"path": wt_path,
"branch": wt_info["branch"],
"track_id": wt_info.get("track_id"),
"track_status": wt_info.get("status"),
"commit": commit,
"short_hash": short_hash,
"is_merged_to_main": is_merged,
"commit_log": log,
"changed_files": [f for f in changed.split("\n") if f],
"patch": diff,
"uncommitted_diff": uncommitted if uncommitted else None,
}
# Files to skip when copying artifacts (large binaries, caches)
_SKIP_EXTENSIONS = {".nsys-rep", ".ncu-rep", ".sqlite", ".qdstrm"}
_SKIP_NAMES = {"__pycache__", ".git"}
_MAX_FILE_SIZE = 200 * 1024 * 1024 # 200 MB
def _copy_ignore(directory: str, files: list) -> list:
"""shutil.copytree ignore callback: skip large binaries and caches."""
ignored = []
for f in files:
if f in _SKIP_NAMES:
ignored.append(f)
continue
fpath = Path(directory) / f
if fpath.suffix in _SKIP_EXTENSIONS:
ignored.append(f)
continue
if fpath.is_file():
try:
if fpath.stat().st_size > _MAX_FILE_SIZE:
ignored.append(f)
except OSError:
pass
return ignored
def snapshot_changes(
artifact_dir: Path,
output_dir: Path,
repo_root: Path,
) -> Dict[str, Any]:
"""Create a full snapshot of all campaign changes.
Returns the manifest dict.
"""
output_dir.mkdir(parents=True, exist_ok=True)
# --- worktrees ---
worktrees = _find_campaign_worktrees(artifact_dir, repo_root)
wt_metadata = []
patches_dir = output_dir / "patches"
patches_dir.mkdir(exist_ok=True)
for wt in worktrees:
snap = _capture_worktree_snapshot(wt)
# Store patch as a separate file (can be large)
safe_name = snap["name"]
if snap.get("patch"):
(patches_dir / f"{safe_name}.patch").write_text(
snap["patch"], encoding="utf-8"
)
if snap.get("uncommitted_diff"):
(patches_dir / f"{safe_name}_uncommitted.patch").write_text(
snap["uncommitted_diff"], encoding="utf-8"
)
# Strip large text from metadata (it's in the files)
meta = {k: v for k, v in snap.items()
if k not in ("patch", "uncommitted_diff")}
meta["has_patch"] = bool(snap.get("patch"))
meta["has_uncommitted"] = bool(snap.get("uncommitted_diff"))
wt_metadata.append(meta)
# --- artifact directory ---
artifacts_dest = output_dir / "campaign_artifacts"
artifacts_copied = False
skipped_files: List[str] = []
if artifact_dir.exists():
shutil.copytree(
artifact_dir, artifacts_dest,
ignore=_copy_ignore, dirs_exist_ok=True,
symlinks=False, # resolve symlinks so we capture actual data
)
artifacts_copied = True
# --- main repo state ---
main_commit = _run_git(["rev-parse", "HEAD"], cwd=str(repo_root)) or "unknown"
main_short = _run_git(["rev-parse", "--short", "HEAD"], cwd=str(repo_root)) or "unknown"
main_diff_stat = _run_git(["diff", "HEAD", "--stat"], cwd=str(repo_root)) or ""
skill_diff = _run_git(
["diff", "HEAD", "--", ".codex/skills/ammo/"], cwd=str(repo_root)
) or ""
if skill_diff:
(output_dir / "skill.patch").write_text(skill_diff, encoding="utf-8")
# --- manifest ---
manifest = {
"snapshot_timestamp": datetime.now(timezone.utc).isoformat(),
"repo_root": str(repo_root),
"artifact_dir": str(artifact_dir),
"main_branch": {
"commit": main_commit,
"short_hash": main_short,
"has_uncommitted_changes": bool(main_diff_stat),
},
"worktrees": wt_metadata,
"worktree_count": len(wt_metadata),
"artifacts_copied": artifacts_copied,
"has_skill_diff": bool(skill_diff),
}
(output_dir / "manifest.json").write_text(
json.dumps(manifest, indent=2) + "\n", encoding="utf-8"
)
return manifest
def main() -> int:
parser = argparse.ArgumentParser(
description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter,
)
parser.add_argument(
"--artifact-dir", type=str, required=True,
help="Path to campaign artifact directory",
)
parser.add_argument(
"--output", type=str, required=True,
help="Output directory for the changes snapshot",
)
parser.add_argument(
"--repo-root", type=str, default=None,
help="Repository root (auto-detected if omitted)",
)
args = parser.parse_args()
artifact_dir = Path(args.artifact_dir).expanduser().resolve()
output_dir = Path(args.output).expanduser().resolve()
if args.repo_root:
repo_root = Path(args.repo_root).expanduser().resolve()
else:
detected = _run_git(["rev-parse", "--show-toplevel"])
if detected:
repo_root = Path(detected)
else:
print("ERROR: Could not detect git repo root. Use --repo-root.",
file=sys.stderr)
return 1
manifest = snapshot_changes(artifact_dir, output_dir, repo_root)
print(f"Snapshot complete: {manifest['worktree_count']} worktrees captured")
print(f"Artifacts copied: {manifest['artifacts_copied']}")
print(f"Output: {output_dir}")
return 0
if __name__ == "__main__":
sys.exit(main())
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |