ericrisco/rsc-harness / skills/roast-me
roast-me
VerifiedUse when someone wants an honest, comedic audit of their own prompting — reads their past agent transcripts, scores prompt quality and compute efficiency, names their worst habits, builds a model-selection cheat sheet. NOT reviewing code an agent produced (that is `code-review`), NOT prompting technique in general (that is `prompt-engineering`).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.4 KB |
| evals/README.md | markdown | 3.8 KB |
| evals/cases.yaml | config | 5.7 KB |
| prompts/analyze.md | markdown | 5.5 KB |
| prompts/compute.md | markdown | 4.8 KB |
| prompts/roast.md | markdown | 7.6 KB |
| tools/adapters/__init__.py | script | 36 B |
| tools/adapters/base.py | script | 2.4 KB |
| tools/adapters/claude.py | script | 4.2 KB |
| tools/adapters/codex.py | script | 8.0 KB |
| tools/adapters/gemini.py | script | 5.4 KB |
| tools/adapters/registry.py | script | 2.0 KB |
| tools/extract_prompts.py | script | 20.2 KB |
tools/adapters/claude.py
View on GitHub"""Claude Code adapter — first-class reference implementation.
Session files: ~/.claude/projects/*/*.jsonl
Format: JSON Lines, one object per line.
Each line is a conversation event with a "type" field:
"user" — user message (prompt or tool result)
"assistant" — model response
User prompt records are lines where:
- type == "user"
- message.isMeta is falsy
- message.content contains at least one text block (not only tool_result blocks)
This is the canonical adapter; its parsing logic is the reference for all others.
Confirmed: path and format verified against live ~/.claude/projects structure on
macOS (2026-06). The JSONL schema is stable across Claude Code versions.
"""
from __future__ import annotations
import json
import time
from pathlib import Path
from typing import Any
from .base import BaseAdapter
class ClaudeAdapter(BaseAdapter):
RUNTIME_ID = "claude"
# Location of Claude Code project session files.
PROJECTS_DIR = Path.home() / ".claude" / "projects"
def discover(self) -> list[Path]:
"""Find all *.jsonl session files under ~/.claude/projects/."""
if not self.PROJECTS_DIR.exists():
return []
files: list[Path] = []
try:
for jsonl in self.PROJECTS_DIR.rglob("*.jsonl"):
try:
if jsonl.is_file():
files.append(jsonl)
except OSError:
continue
except OSError:
return []
return sorted(files, key=lambda f: f.stat().st_mtime, reverse=True)
def parse(self, path: Path) -> list[dict[str, Any]]:
"""Parse one Claude Code .jsonl session file."""
records: list[dict[str, Any]] = []
try:
lines = path.read_text(encoding="utf-8", errors="replace").splitlines()
except OSError:
return []
for line in lines:
line = line.strip()
if not line:
continue
try:
obj = json.loads(line)
except json.JSONDecodeError:
continue
if obj.get("type") != "user":
continue
msg = obj.get("message", {})
if msg.get("isMeta"):
continue
content = msg.get("content", [])
text = _extract_text(content)
if not text.strip():
continue
# Skip messages that are only tool_result blocks
if _is_only_tool_results(content):
continue
ts = _parse_timestamp(obj)
records.append({
"runtime": self.RUNTIME_ID,
"session_file": str(path),
"prompt_text": text,
"timestamp": ts,
# Forward the raw object so the main extractor can inspect
# adjacent lines for error/correction context.
"_raw": obj,
})
return records
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
def _extract_text(content: Any) -> str:
if isinstance(content, str):
return content
if isinstance(content, list):
parts = []
for block in content:
if isinstance(block, dict) and block.get("type") == "text":
parts.append(block.get("text", ""))
return "\n".join(parts)
return ""
def _is_only_tool_results(content: Any) -> bool:
if not isinstance(content, list):
return False
has_text = any(
isinstance(b, dict) and b.get("type") == "text" for b in content
)
has_tool_result = any(
isinstance(b, dict) and b.get("type") == "tool_result" for b in content
)
return has_tool_result and not has_text
def _parse_timestamp(obj: dict) -> float | None:
ts = obj.get("timestamp")
if ts is None:
return None
if isinstance(ts, (int, float)):
return float(ts)
if isinstance(ts, str):
try:
import datetime
dt = datetime.datetime.fromisoformat(ts.replace("Z", "+00:00"))
return dt.timestamp()
except (ValueError, AttributeError):
pass
return None
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |