Commit Graph
3 Commits
Author SHA1 Message Date
Willie 5d9565098e fix(audit): close regex bypass vulnerabilities and improve code reuse
Security fixes from red-team review of audit rules:
- Prompt injection: add \s* after ^ anchor to prevent whitespace bypass,
  add (?i) for case-insensitive matching, use [^>]* for HTML attributes,
  expand to cover DEV MODE / DEVELOPER_MODE variants
- Credential access: add rules for ln/cp symlink, input redirection,
  dd if=/etc/shadow (previously mislabeled as destructive-commands)
- Data exfiltration: remove exclude pattern that allowed .png?stolen_data
- Interpreter tier: add bun, deno, npx, tsx, pwsh, powershell; add
  version-suffix stripping (python3.11 → python3); handle env prefix

Code quality (from /simplify review):
- Rename crossFinding → syntheticFinding and reuse in TierCombinationFindings
  to eliminate 5 inline Finding literals
- Extract hasFinding() test helper to reduce tier test boilerplate
- Remove unnecessary (?i) from credential-access-3/4 (Unix commands are
  always lowercase)
2026-03-03 16:08:11 +08:00
Willie 7a6ba4546d feat(audit): expand prompt injection detection, add interpreter tier and credential rules
Strengthen the audit engine with new detection capabilities inspired by
skilld's sanitization research and Snyk ToxicSkills findings:

- Expand prompt-injection-1 to detect OVERRIDE:/IGNORE:/ADMIN:/ROOT: prefixes
- Add prompt-injection-2: agent directive XML tags (<system>, <instructions>, etc.)
- Add prompt-injection-3: DEVELOPER MODE, JAILBREAK, DAN MODE directives
- Add data-exfiltration-2: external images with query params (exfil vector)
- Add credential-access-3/4: /etc/shadow, /etc/passwd and related system files
- Add TierInterpreter (T6): classify python/node/ruby/perl/lua/php as
  Turing-complete runtimes separate from build tools
- Add tier-interpreter and tier-interpreter-network combination findings
- Add cross-skill-cred-interpreter rule for credential × interpreter pairs
- Update docs to reflect 38 rules and new tier/cross-skill patterns
2026-03-03 16:08:11 +08:00
Willie 4d17ff5d57 feat: add security audit — scan engine, CLI, install hooks, web UI, and docs
Scan skills for prompt injection, data exfiltration, credential access,
destructive commands, obfuscation, and suspicious URLs. CRITICAL findings
block installation by default; HIGH/MEDIUM shown as warnings.

- Scan engine: internal/audit with regex-based pattern matching
- CLI: `skillshare audit [name]` with per-skill progress and summary
- Install hooks: auto-scan after copy, block on CRITICAL unless --force
- Web UI: AuditPage with run/rescan, InstallForm shows ConfirmDialog
  for audit blocks (force-retry) and warning details
- ConfirmDialog: add `wide` prop, hide cancel when cancelText is empty
- Reduced false positives: rm -rf only flags root/glob, skip localhost
  URLs, exclude `fetch` as English word, hidden-unicode downgraded to HIGH
- README: add audit section, update screenshots, unify image sizes,
  add audit to Why/Comparison/Daily Commands
- Docs: audit command page, install security scanning section, FAQ,
  common errors, daily workflow, sidebar, landing page, intro
- Blog specs: add audit sections to Dev.to and Medium drafts
2026-02-10 01:53:55 +08:00