Please report security issues privately. Do not open a public GitHub issue for vulnerabilities that could put people or data at risk.
- Preferred: GitHub Security Advisories ("Report a vulnerability") on this repository.
- Alternative: contact the maintainer via the email on the GitHub profile,
with the subject line
[TRACEFIELD security].
Include: affected version/commit, reproduction steps, impact, and any known workarounds. You will get an acknowledgement when the maintainer is available; there is no published response SLA — this is a small project and a fake SLA would be dishonest. You will be credited in the changelog unless you ask not to be.
TRACEFIELD is a local-first tool: your case data stays on your machine. That shapes the threat model:
In scope
- Path traversal or arbitrary file read/write via crafted case IDs, evidence filenames, or export paths.
- SSRF bypasses: any URL that makes
tracefield source add --fetch(or the username checker) connect to loopback, private, link-local, multicast, or cloud-metadata addresses, including via DNS rebinding or redirects. - Terminal escape/ANSI injection that can hide or forge output when rendering untrusted source text, filenames, or tool output.
- Secret leakage: redaction bypasses, evidence/export paths that write secrets un-redacted, or environment-variable values appearing in logs/reports.
- Integrity failures: cases where
evidence verifyorcase-verifyreports OK for tampered bytes, or where the event hash chain verifies after a row was modified. - Dependency vulnerabilities with a demonstrated path to any of the above.
Out of scope (by design — not bugs)
- The event hash chain is tamper-evident within this store's threat model only; a sufficiently privileged operator who rewrites every row can defeat it. This is documented in the code ("Not a blockchain").
- SHA-256 evidence digests prove byte-level integrity, not truthfulness, authorship, or authenticity of content.
- Third-party tools and providers reachable through opt-in network features (their own vulnerabilities belong with their maintainers).
- Findings that require physical or root access to the machine already holding the case database.
- Denial of service against yourself (e.g., attaching a pathological file to your own case).
- Network is off by default;
TRACEFIELD_ALLOW_NETWORK=1plus an explicit--fetch/check flag are both required, and the SSRF policy is re-evaluated per request and per redirect. - At rest: data directories
0700, database0600, evidence copies0400. - All external strings are sanitized before terminal rendering; JSON output
is escaped by
json.dumps. - Catalogue content is inert data; nothing in
registry/may import or execute catalogue-supplied content. - CI runs the security regression suite (
pytest -m security) andpip-audit --strict; a red gate is a signal, not noise.
Only the latest release on main receives security fixes. There are no
long-term support branches at this stage of the project.