Skip to content

Security: bitflicker64/Termstory

Security

SECURITY.md

Security Policy

Supported Versions

Version Supported
0.6.x ✅
< 0.6 ❌

Reporting a Vulnerability

If you discover a security issue in TermStory, do not open a public GitHub issue — this gives potential attackers advance notice before a fix is available.

Instead, use GitHub's private security advisory system:

  1. Navigate to the repository Security tab.
  2. Click "Report a vulnerability".
  3. Describe the issue, steps to reproduce, and potential impact.

Expected response time: initial acknowledgement within 48 hours, fix within 14 days for confirmed vulnerabilities.

Trust Model and Known Limitations

TermStory's privacy guarantee is: no raw data leaves your machine without passing through termstory/sanitizer.py first. Understanding the limits of that guarantee is important.

What the sanitizer does well

  • Named-prefix secrets (AWS AKIA*, OpenAI sk-*, Anthropic sk-ant-*, Slack xoxb-*, Google AIzaSy*, etc.) are caught by specific regex patterns.
  • Structural patterns (--password=, export KEY=***) are redacted regardless of the specific value.
  • Blacklisted workflows (vault, aws configure, gh auth, raw token strings, kubectl create secret) gate the entire COMMANDS block — none of those raw command strings reach the LLM.

Known gaps (by design — documented here, not hidden)

Limitation Detail
Short / low-entropy secrets redact_high_entropy() requires ≥ 24 characters and Shannon entropy > 4.3 bits/char. A short or predictable secret with no recognizable prefix (e.g., a 12-character password) will not be flagged.
Novel token formats Blacklist and named-pattern detection is structural. A company-internal token format with no known prefix silently passes through unless you add it to ~/.termstoryignore.
Commit message prose redact_command() was designed for shell syntax. A secret embedded in natural-language commit text with no =, flag, or prefix cue may not be detected.
.termstoryignore not live-reloaded Custom patterns load once at process startup. Edits require a TermStory restart to take effect.
Model refusal is not a control TermStory adds an instruction in termstory ask prompts asking the LLM never to output credentials verbatim. This is a secondary layer; the redaction pass is the actual control.

Recommended mitigations

  • Use HISTCONTROL=ignorespace in your shell and prefix any command containing a plaintext secret with a space — the shell will refuse to write it to history, and TermStory will never see it.
  • Use Ollama for fully local inference if you work with highly sensitive codebases.
  • Maintain ~/.termstoryignore with patterns for any proprietary token formats your workflows produce.

There aren't any published security advisories