Aftermath executes repository commands to produce verification evidence. Treat it as a privileged local tool.
| Version | Supported |
|---|---|
| 0.4.x | Yes |
| 0.3.x | Yes |
| 0.2.x | Best-effort |
| 0.1.x | Best-effort |
- Runs build/test/lint/typecheck/smoke commands discovered from configuration, CI, package scripts, and ecosystem conventions
- Captures Git diffs, logs, and receipts on disk under
.aftermath/ - Optionally exposes results through a local MCP server
| Threat | Mitigation |
|---|---|
| Malicious repository scripts | Prefer explicit .aftermath.toml; never invent scripts; flag destructive commands |
| Command injection | Commands are executed with spawn + argv arrays (no shell) |
| Poisoned configuration | Validate/limit discovered commands; destructive patterns require approval |
| Secret leakage in logs | Best-effort redaction of common token/key patterns (not perfect) |
| Oversized output | Log size limits and truncation metadata |
| Privilege escalation | No sudo, no global package-manager mutation, no publish/deploy/push by default |
| Process escape / unexpected tools | Working directory scoped; timeouts; cancellation |
Do not run Aftermath against untrusted repositories without reviewing:
.aftermath.tomlpackage.json/Cargo.toml/ CI workflows that may be discovered as commands- pre/post scripts that tooling may invoke indirectly
- The Cursor plugin should be reviewed before installation
- The MCP server exposes deterministic local data and verification entrypoints
- No cloud backend, no required API keys, no telemetry in v1
Receipts and logs may contain source paths, failure output, and accidentally printed secrets. Redaction is best-effort. Do not commit bulky run logs. Prefer committing only baseline.json and policy/config when desired.
Please report security issues privately via GitHub Security Advisories on this repository:
https://github.com/theworker02/aftermath/security/advisories/new
If advisories are unavailable, contact the maintainer through the GitHub profile for theworker02 (prefer private channels).
Do not open public issues for undisclosed vulnerabilities.
We aim to acknowledge reports within 7 days and ship fixes as quickly as practical for confirmed issues.