logomesh

Paste a Sentry URL. Get a failing pytest back.

pip install logomesh
logomesh repro https://sentry.io/organizations/your-org/issues/12345678/

Why logomesh

• Deterministic test synthesis — no LLM touches the test bytes. The pytest comes from a pure function of your frame locals.
• Verified reproduction — the sandbox must raise the same exception type Sentry captured. If it doesn't, it refuses to ship.
• Runs on your machine — your Sentry token, your Docker, your repo. Nothing phones home.
• PII redaction built in — strips card numbers, emails, JWTs, API keys, and sensitive field names before anything hits an LLM.
• Audit-ready — every artifact carries llm_in_evidence_path: false, a SHA-256 stamp, and SOC2/PCI control mapping.

Requirements

• Python 3.11+
• Docker running locally
• Sentry auth token with event:read scope (Settings → API Keys)
• OpenAI API key (powers the agent — context recovery, dep pinning, source resolution, retries)

export SENTRY_AUTH_TOKEN=sntryu_...
export OPENAI_API_KEY=sk-...

Or drop them in a .env file in your project root.

Usage

logomesh repro <sentry-url>                  # reproduce a crash
logomesh repro <url> --no-llm               # deterministic only, skip LLM
logomesh repro <url> --artifact             # emit SOC2/PCI audit artifact
logomesh repro <url> --draft-pr             # open GitHub draft PR with failing test
logomesh repro <url> --json                 # machine-readable output
logomesh repro <url> --repo /path/to/repo   # point at local repo (default: cwd)
logomesh repro <url> --timeout 120          # wall-clock timeout in seconds

Supported Sentry URL formats:

https://sentry.io/organizations/{org}/issues/{id}/
https://sentry.io/issues/{id}/
https://{org}.sentry.io/issues/{id}/

Example output

  ✓ Reproduced: ZeroDivisionError at billing/calc.py:18
     division by zero
     rate = total / count

  ✗ Cannot reproduce ValueError at checkout.py:42
     The synthesized test passed against the current branch.
     Either the bug is fixed, or the captured locals are insufficient.

How it works

1. Fetches the innermost in-app frame + locals from your Sentry event
2. PII redaction runs before any data touches an LLM
3. Builds a pytest deterministically from the captured locals
4. Runs it in an airgapped Docker sandbox (128 MB, no network, nobody user)
5. Verifies the sandbox raised the same exception type Sentry captured
6. Emits the failing test — or a structured reason it couldn't reproduce

A LangGraph agent handles context recovery, dep pinning, source resolution, and retries. It's advisory — it never writes the test or the artifact.

Full architecture details → GitHub

What reproduces well

• Input validation bugs, NoneType mismatches, type coercion errors
• Off-by-one, ordering, idempotency issues
• Anything where the crash inputs are captured in the Sentry frame

What doesn't

• Race conditions, distributed failures, live DB/Redis state not in the frame
• C extension crashes, timezone/DST edge cases (sandbox runs TZ=UTC)

When it can't reproduce, it says so with a structured reason. Never guesses.

Sentry setup

Frame locals must be enabled:

sentry_sdk.init(dsn="...", send_default_pii=True)

Or: Project Settings → SDK Setup → Enable "Send default PII"

License

MIT