Pixel-CLI/pixel / .agents/skills/arena
arena
VerifiedRun one of eight standard prompts in two Codex panes side by side in Herdr, one with Pixel's prompt-submit brief and one without, then compare tool calls, answers and wall time. One command runs, one aborts.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.0 KB |
| arena.shexecutable | script | 55.5 KB |
| holdout.tsv | other | 1.3 KB |
| holdout2.tsv | other | 1.3 KB |
| hook-wrap.pyexecutable | script | 5.1 KB |
| offline-protocol-test.pyexecutable | script | 6.7 KB |
| session-digest.pyexecutable | script | 4.6 KB |
| single.tsv | other | 1.3 KB |
session-digest.py
View on GitHub#!/usr/bin/env python3
# SPDX-FileCopyrightText: The Pixel contributors
# SPDX-License-Identifier: MIT
"""Split one Claude Code session transcript into turns.
A turn starts at a human prompt and holds what the UserPromptSubmit hooks added (the
brief), the tool calls the agent made, and its last text. Used by `arena.sh run-set
--keep-context`, where one session answers every question of a set.
session-digest.py FILE all turns, JSON
session-digest.py FILE --turn N turn N (1-based), JSON
session-digest.py FILE --hide-prompt the prompt is replaced by its sha256 prefix
session-digest.py A.jsonl --md B.jsonl a Markdown report of the two arms, turn by turn
"""
import hashlib
import json
import sys
def blocks(message):
content = (message or {}).get("content")
return content if isinstance(content, list) else []
def prompt_text(entry):
content = (entry.get("message") or {}).get("content")
if isinstance(content, str):
return content
parts = blocks(entry.get("message"))
if not parts or any(b.get("type") == "tool_result" for b in parts):
return None
return "".join(b.get("text", "") for b in parts if b.get("type") == "text") or None
def call_of(block):
args = block.get("input") or {}
for key in ("command", "file_path", "pattern", "query", "path"):
if args.get(key):
return str(args[key])
return json.dumps(args, sort_keys=True)
def turns(path, hide_prompt=False):
out, cur = [], None
with open(path, encoding="utf-8", errors="replace") as handle:
for line in handle:
try:
entry = json.loads(line)
except ValueError:
continue
kind = entry.get("type")
if kind == "user" and not entry.get("isMeta"):
text = prompt_text(entry)
if text:
digest = hashlib.sha256(text.encode()).hexdigest()[:16]
cur = {"n": len(out) + 1, "prompt": f"sha256:{digest}" if hide_prompt else text,
"timestamp": entry.get("timestamp"), "briefs": [], "tool_calls": [], "answer": None}
out.append(cur)
elif cur is None:
continue
elif kind == "attachment":
att = entry.get("attachment") or {}
if att.get("type") == "hook_additional_context" and att.get("hookEvent") == "UserPromptSubmit":
body = att.get("content")
body = "\n".join(body) if isinstance(body, list) else str(body or "")
if body.strip():
cur["briefs"].append(body)
elif kind == "assistant":
for block in blocks(entry.get("message")):
if block.get("type") == "tool_use":
cur["tool_calls"].append({"name": block.get("name"), "call": call_of(block)})
elif block.get("type") == "text" and block.get("text", "").strip():
cur["answer"] = block["text"]
return out
def markdown(pixel, nopixel):
lines = ["# Session report: every brief and tool call, turn by turn", ""]
for i in range(max(len(pixel), len(nopixel))):
p = pixel[i] if i < len(pixel) else None
n = nopixel[i] if i < len(nopixel) else None
head = (p or n)["prompt"]
lines += [f"## Turn {i + 1}", "", f"> {head}", ""]
if p and n and p["prompt"] != n["prompt"]:
lines += [f"**The two arms answered different prompts in this turn.** Without Pixel: {n['prompt']}", ""]
for label, t in (("with Pixel", p), ("without Pixel", n)):
if t is None:
continue
lines += [f"### {label}: {len(t['tool_calls'])} tool call(s), brief {sum(len(b.encode()) for b in t['briefs'])} bytes", ""]
for b in t["briefs"]:
lines += ["<details><summary>brief</summary>", "", "```", b, "```", "</details>", ""]
for c in t["tool_calls"]:
lines.append(f"- `{c['name']}` {c['call'][:200]}")
lines += ["", "answer:", "", (t["answer"] or "(none)"), ""]
return "\n".join(lines)
def main(argv):
if len(argv) < 2:
sys.exit(__doc__)
path, hide = argv[1], "--hide-prompt" in argv
if "--md" in argv:
other = argv[argv.index("--md") + 1]
print(markdown(turns(path, hide), turns(other, hide)))
return
result = turns(path, hide)
if "--turn" in argv:
n = int(argv[argv.index("--turn") + 1])
result = result[n - 1] if 0 < n <= len(result) else None
print(json.dumps(result, indent=1))
if __name__ == "__main__":
main(sys.argv)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 467k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |