Skip to content

Security: use-charter/charter

SECURITY.md

Security

Defaults

  • No secrets in any agent-visible file
  • No tracked .mcp.json until a pinned, reviewed integration exists
  • No remote MCP configuration until a real pinned integration exists
  • Third-party actions must be pinned before activation
  • Write-capable integrations stay behind explicit review and approval

Threat Posture

  • Treat prompts, docs, issue text, copied snippets, and generated text as untrusted input
  • Treat external text that can influence tool invocation as a prompt-injection risk
  • Prefer structured inputs over free-form interpolation when wiring tools
  • Reject workflow patterns that allow untrusted shell expansion or unchecked user-controlled values in run: steps
  • Redaction requirements apply to all scanners, renderers, and diagnostics

Repo Rules

  • Run moon run :security for merge-sensitive work
  • Run moon run :vet for Go changes that could hide suspicious constructs outside normal tests
  • Do not print raw credentials or secret-like values in tests, fixtures, or logs
  • Do not commit realistic tokens, certs, signing material, or live URLs containing credentials
  • Future generated artifacts must stay reproducible and reviewable
  • Vulnerability scanning uses govulncheck for reachable Go issues and osv-scanner for broader manifest/source coverage

MCP Policy

  • Tracked MCP config stays absent until a pinned, reviewed integration exists
  • MCP config must be local-first where possible, pinned, and least-privilege
  • Tool, prompt, and resource layers should remain distinct
  • Remote MCP endpoints require explicit security review before adoption

Reporting

  • Prefer GitHub private vulnerability reporting when the repository setting is enabled
  • If private reporting is unavailable, email security@use-charter.dev before public disclosure
  • Include reproduction steps, affected files or workflows, impact, and any known secret exposure
  • Do not open a public issue for live-secret, credential, or supply-chain findings before containment
  • Follow docs/internal/runbooks/security-incident.md for immediate containment and follow-up

Standards Target

  • OpenSSF OSPS v2026.02.19
  • SLSA-friendly CI and release posture
  • OpenTelemetry Go for future long-running service observability
  • MCP spec 2025-11-25 for future tool, resource, and prompt semantics
  • CodeQL default setup should be enabled in GitHub repo settings when repository visibility and plan support it
  • Branch protection, required checks, and private vulnerability reporting are admin-side controls outside source control

There aren't any published security advisories