Pixel-CLI/pixel / .agents/skills/pixel-retro
pixel-retro
VerifiedMine the last 24 h (or a given window) of pixel usage — every repo's `.pixel/actions.jsonl` plus the agent transcripts indexed by `pixel recall` (claude, codex, pi…) — for frictions pixel actually caused, verify each one against the current binary and code, and propose ranked, evidence-backed improvements (bug fix, error message, flag, agent prompt, rule, doc); for the pixel repo itself, also measure each pull request from first edit to validated CI and where agents sat idle (`lead_time.py`). Suggests only; implements nothing without the user's pick. Use when the user says "/pixel-retro", "retro pixel", "qu'est-ce qui a coincé avec pixel", "améliorations pixel depuis les transcripts", "pixel friction", "où l'agent attend", "bottleneck", "lead time", or asks what to improve in pixel or its development workflow from recent sessions.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 21.9 KB |
| adherence.py | script | 18.3 KB |
| lead_time.py | script | 22.7 KB |
| ledger.py | script | 1.5 KB |
| roots.py | script | 3.2 KB |
| slow.py | script | 5.6 KB |
| test_adherence.py | script | 9.6 KB |
| test_lead_time.py | script | 17.4 KB |
| test_slow.py | script | 2.4 KB |
test_adherence.py
View on GitHub#!/usr/bin/env python3
# SPDX-FileCopyrightText: The Pixel contributors
# SPDX-License-Identifier: MIT
"""Contract of adherence.py: which call counts as what, and which numbers come out.
Its numbers decide whether a hook, prompt or routing change moved agent
behaviour, so each case pins a way to get them wrong: a `| grep` filter on
Pixel output counted as a native search, a bounded `sed -n` counted as an
unbounded read, a Codex edit inside an `exec` script missed, hosts pooled,
a session outside an indexed repository counted.
"""
import datetime
import json
import pathlib
import sys
import tempfile
import unittest
sys.path.insert(0, str(pathlib.Path(__file__).parent))
import adherence as ad # noqa: E402
T0 = datetime.datetime(2026, 10, 1, 10, 0, tzinfo=datetime.timezone.utc)
def at(minutes):
return (T0 + datetime.timedelta(minutes=minutes)).isoformat().replace("+00:00", "Z")
def claude_use(minutes, name, inp, cwd):
return json.dumps({"type": "assistant", "timestamp": at(minutes), "cwd": cwd,
"message": {"content": [{"type": "tool_use", "id": f"t{minutes}", "name": name,
"input": inp}]}})
def codex_exec(minutes, script, cwd):
return json.dumps({"timestamp": at(minutes), "type": "response_item",
"payload": {"type": "custom_tool_call", "name": "exec", "input": script, "cwd": cwd}})
class ShellClassification(unittest.TestCase):
def test_pixel_retrieval_is_pixel_and_its_filter_is_not_a_search(self):
self.assertEqual(ad.shell_events("pixel search-content -F foo | grep bar"), [("pixel", "search-content")])
self.assertEqual(ad.shell_events("rtk pixel find-code 'x' 2>&1 | head -40"), [("pixel", "find-code")])
self.assertEqual(ad.shell_events("pixel status"), [], "status is not retrieval")
def test_quoted_text_and_heredoc_bodies_are_not_commands(self):
self.assertEqual(ad.shell_events('echo "noise; pixel impact X; rg foo"'), [])
self.assertEqual(ad.shell_events("echo 'a | grep b' && rg c"), [("search", "rg")])
heredoc = "cat <<'EOF' > notes.md\npixel impact Foo\nrg needle\nEOF\nrg after"
self.assertEqual(ad.shell_events(heredoc), [("search", "rg")])
def test_native_searches_in_any_statement(self):
self.assertEqual(ad.shell_events("cd /repo && rg -n foo src; git grep bar"),
[("search", "rg"), ("search", "git")])
self.assertEqual(ad.shell_events("find . -name '*.rs'"), [("search", "find")])
self.assertEqual(ad.shell_events("find . -type f -newer x"), [], "a find without a name test lists")
def test_read_widths(self):
self.assertEqual(ad.shell_events("sed -n '10,49p' a.rs"), [("read", 40)])
self.assertEqual(ad.shell_events("sed -n '10,+40p' a.rs"), [("read", 41)])
self.assertEqual(ad.shell_events("sed -n 12p a.rs"), [("read", 1)])
self.assertEqual(ad.shell_events("sed -n '5,$p' a.rs"), [("read", None)])
self.assertEqual(ad.shell_events("sed -i s/a/b/ a.rs"), [], "an in-place edit is not a read")
self.assertEqual(ad.shell_events("cat a.rs"), [("read", None)])
self.assertEqual(ad.shell_events("head -n 30 a.rs"), [("read", 30)])
self.assertEqual(ad.shell_events("tail a.rs"), [("read", 10)])
self.assertEqual(ad.shell_events("head -n -40 a.rs"), [("read", None)], "all but the last 40")
self.assertEqual(ad.shell_events("tail -n +40 a.rs"), [("read", None)], "from line 40 on")
self.assertEqual(ad.shell_events("tail -n 40 a.rs"), [("read", 40)])
self.assertEqual(ad.shell_events("head -25 a.rs"), [("read", 25)])
self.assertEqual(ad.shell_events("cat"), [], "no operand reads stdin")
def test_claude_tools(self):
self.assertEqual(ad.claude_events("Read", {"file_path": "a", "limit": 40}), [("read", 40)])
self.assertEqual(ad.claude_events("Read", {"file_path": "a", "offset": 10}), [("read", None)])
self.assertEqual(ad.claude_events("Grep", {}), [("search", "Grep")])
self.assertEqual(ad.claude_events("MultiEdit", {}), [("edit", None)])
self.assertEqual(ad.claude_events("mcp__pixel__find_code", {}), [("pixel", "find-code")])
self.assertEqual(ad.claude_events("mcp__pixel__who_calls", {}), [("pixel", "who-calls")])
self.assertEqual(ad.claude_events("mcp__pixel__status", {}), [], "status is not retrieval")
def test_codex_exec_scripts_keep_call_order(self):
script = ('const a = await tools.exec_command({"cmd":"rg -n foo src","workdir":"/r"});\n'
'await tools.apply_patch("*** Begin Patch");\n'
'const b = await tools.exec_command({"cmd":"pixel impact \\"Foo::bar\\"","workdir":"/r"});')
self.assertEqual(ad.exec_script_events(script),
[("search", "rg"), ("edit", None), ("pixel", "impact")])
def test_each_codex_call_reads_its_own_command(self):
# An unquoted key on the first call must not borrow the next call's command.
script = ('await tools.exec_command({cmd: "rg -n foo src"});\n'
'await tools.apply_patch("x");\n'
'await tools.exec_command({"cmd": "pixel impact Bar"});')
self.assertEqual(ad.exec_script_events(script),
[("search", "rg"), ("edit", None), ("pixel", "impact")])
class Summary(unittest.TestCase):
def test_counts_per_session_stream(self):
sessions = [
# Pixel first, then a native search: the answer was not trusted.
[("pixel", "search-content"), ("search", "rg"), ("read", None), ("edit", None)],
# Native first; a structural Pixel call before the first edit.
[("search", "Grep"), ("pixel", "impact"), ("read", 40), ("edit", None)],
# No search at all.
[("read", 200)],
]
s = ad.summarize(sessions)
self.assertEqual(s["sessions"], 3)
self.assertEqual((s["pixel_calls"], s["native_searches"]), (2, 2))
self.assertEqual(s["pixel_share"], 0.5)
self.assertEqual((s["first_search_pixel"], s["sessions_with_search"]), (1, 2))
self.assertEqual(s["sessions_with_pixel"], 2)
self.assertEqual(s["pixel_then_native"], 1, "of 2 Pixel calls")
self.assertEqual((s["reads"], s["unbounded_reads"], s["wide_reads"]), (3, 1, 1))
# The read after `search-content` follows a native search, so it is not
# a read after Pixel; the 40-line read after `impact` is.
self.assertEqual((s["reads_after_pixel"], s["unbounded_after_pixel"]), (1, 0))
self.assertEqual(s["median_read_width_after_pixel"], 40)
self.assertEqual((s["impact_before_edit"], s["edit_sessions"]), (1, 2))
def test_after_pixel_means_the_very_next_event(self):
s = ad.summarize([[("pixel", "find-code"), ("edit", None), ("search", "rg"), ("read", None)]])
self.assertEqual(s["pixel_then_native"], 0, "an edit came in between")
self.assertEqual(s["reads_after_pixel"], 0)
s = ad.summarize([[("pixel", "find-code"), ("read", 30), ("pixel", "impact"), ("search", "Grep")]])
self.assertEqual((s["reads_after_pixel"], s["pixel_then_native"]), (1, 1))
def test_empty_host_reports_no_share(self):
s = ad.summarize([])
self.assertIsNone(s["pixel_share"])
self.assertIsNone(s["median_read_width_after_pixel"])
class Collect(unittest.TestCase):
def test_hosts_are_separate_and_unindexed_repositories_skipped(self):
with tempfile.TemporaryDirectory() as tmp:
tmp = pathlib.Path(tmp)
indexed, plain = tmp / "indexed", tmp / "plain"
for repo in (indexed, plain):
(repo / ".git").mkdir(parents=True)
(indexed / ".pixel").mkdir()
projects = tmp / "claude"
(projects / "p").mkdir(parents=True)
(projects / "p" / "s1.jsonl").write_text("\n".join([
claude_use(1, "Bash", {"command": "pixel search-content -F foo"}, str(indexed)),
claude_use(2, "Grep", {"pattern": "foo"}, str(indexed)),
]))
(projects / "p" / "s2.jsonl").write_text(claude_use(1, "Grep", {"pattern": "x"}, str(plain)))
# Active in the window, in the indexed repository, with no retrieval at all.
(projects / "p" / "s3.jsonl").write_text(claude_use(3, "TodoWrite", {}, str(indexed)))
codex = tmp / "codex" / "2026" / "10" / "01"
codex.mkdir(parents=True)
(codex / "rollout-a.jsonl").write_text("\n".join([
json.dumps({"timestamp": at(0), "type": "session_meta", "payload": {"cwd": str(indexed)}}),
codex_exec(1, 'await tools.exec_command({"cmd":"rg foo"})', str(indexed)),
]))
start = T0 - datetime.timedelta(hours=1)
result = ad.collect(start, [projects], tmp / "codex", all_repos=False)
self.assertEqual(result["claude"]["sessions"], 2,
"the plain repository is not counted; an active session without retrieval is")
self.assertEqual(result["claude"]["sessions_with_pixel"], 1)
self.assertEqual((result["claude"]["pixel_calls"], result["claude"]["native_searches"]), (1, 1))
self.assertEqual((result["codex"]["pixel_calls"], result["codex"]["native_searches"]), (0, 1))
self.assertEqual(ad.collect(start, [projects], tmp / "codex", all_repos=True)["claude"]["sessions"], 3)
later = T0 + datetime.timedelta(minutes=5)
self.assertEqual(ad.collect(later, [projects], tmp / "codex", all_repos=True)["claude"]["sessions"], 0,
"events before the window start are not counted")
if __name__ == "__main__":
unittest.main(verbosity=1)
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: 466k | Updated | Claude CodeCodex | Verified | 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: 466k | 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 |