fix(skills): update trailofbits/skills to cfe5d7b, fix zeroize-audit - #740
Closed
JAORMX wants to merge 3 commits into
Closed
fix(skills): update trailofbits/skills to cfe5d7b, fix zeroize-audit#740JAORMX wants to merge 3 commits into
JAORMX wants to merge 3 commits into
Conversation
…onstant-time-analysis,differential-review,fp-check,insecure-defaults,property-based-testing,sarif-parsing,semgrep,semgrep-rule-creator,semgrep-rule-variant-creator,sharp-edges,supply-chain-risk-auditor,variant-analysis,yara-rule-authoring,zeroize-audit
Scanner flagged two list-form subprocess.run() calls (generate_poc.py, check_rust_asm.py) as dangerous. Neither uses shell=True or takes attacker-controlled arguments -- fixed commands/scripts, one with an explicit timeout. Note: the other 5 skills bumped in this digest (agentic-actions-auditor, codeql, constant-time-analysis, sharp-edges, yara-rule-authoring) still fail skill-security-scan due to the known scanner meta-analyzer JSON-parse bug (dozens to hundreds of raw findings surviving as blocking) -- not fixed here, not practically allowlistable by hand. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Contributor
🛡️ Skill Security Scan Results✅ agentic-actions-auditor
❌ codeql
Blocking issues:
❌ constant-time-analysis
Blocking issues:
✅ differential-review
✅ fp-check
✅ insecure-defaults
✅ property-based-testing
✅ sarif-parsing
✅ semgrep
✅ semgrep-rule-creator
✅ semgrep-rule-variant-creator
❌ sharp-edges
Blocking issues:
Allowlisted (not blocking):
✅ supply-chain-risk-auditor
✅ variant-analysis
❌ yara-rule-authoring
Blocking issues:
✅ zeroize-audit
Summary: Scanned 16 skill(s), found 647 blocking issue(s).
|
samuv
approved these changes
Jul 3, 2026
Contributor
|
#855 advances the trailofbits skills beyond this ref. The later |
danbarr
added a commit
that referenced
this pull request
Aug 25, 2026
…oize-audit Same finding, same false positive already verified in #740: both subprocess.run() calls in tools/generate_poc.py and tools/scripts/check_rust_asm.py are list-form, use fixed commands/scripts, and take no attacker-controlled arguments. Revalidated against the current ref before carrying forward. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Contributor
|
Superseded by #855, which has now carried forward the |
1 task
danbarr
added a commit
that referenced
this pull request
Aug 25, 2026
* chore(deps): update trailofbits/skills digest to 7b9bd5f * chore(skills): bump spec.version for agentic-actions-auditor,codeql,constant-time-analysis,differential-review,fp-check,insecure-defaults,property-based-testing,sarif-parsing,semgrep,semgrep-rule-creator,semgrep-rule-variant-creator,sharp-edges,supply-chain-risk-auditor,variant-analysis,yara-rule-authoring,zeroize-audit * chore: allowlist scanner false positive(s) for supply-chain-risk-auditor * fix(skills): remove insecure-defaults, restructured out of existence upstream trailofbits/skills dropped the skills/ subdirectory from the insecure-defaults plugin at the new digest, so plugins/insecure-defaults/ no longer contains a SKILL.md anywhere. There is nothing left to package at the vendored path. Already-published insecure-defaults artifacts are unaffected. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix(skills): accept LLM_PROMPT_INJECTION risk for semgrep ruleset cloning The scanner's prompt-injection finding on skills/semgrep is not a keyword false positive: the skill genuinely clones third-party ruleset repos by GitHub URL. Accepting as a known, bounded risk since the ruleset sources are fixed and skill-selected (not user-supplied), the content is declarative YAML rather than executable code, and the user approves the exact ruleset list before any clone happens. Same reasoning already accepted for skills/hf-mcp and skills/huggingface-tool-builder. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix(skills): allowlist LLM_COMMAND_INJECTION false positive on semgrep The scanner flagged [TARGET]/[OUTPUT_DIR] placeholders in the scanner subagent's bash templates as unsanitized command injection. Both are locally-resolved values (the codebase path being scanned, and the skill's own output directory) that the user/agent already controls, not attacker-supplied input. Same reasoning already accepted for skills/huggingface-tool-builder and skills/huggingface-paper-publisher. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix(skills): carry forward BEHAVIOR_EVAL_SUBPROCESS allowlist for zeroize-audit Same finding, same false positive already verified in #740: both subprocess.run() calls in tools/generate_poc.py and tools/scripts/check_rust_asm.py are list-form, use fixed commands/scripts, and take no attacker-controlled arguments. Revalidated against the current ref before carrying forward. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix(skills): allowlist LLM_COMMAND_INJECTION findings on codeql Two blocking findings in workflows/build-database.md: - Tracing compilation of a compiled-language codebase requires running that codebase's own build system (make/cmake/gradle/etc.), including a make --dry-run compile-command extraction for the macOS arm64e workaround. This is CodeQL's documented, required mechanism for compiled-language analysis, not a new injection primitive - the same inherent risk as running any CI/build pipeline against the analyzed codebase. - $CMD/$DB_NAME/$OUTPUT_DIR in the bash templates are the skill's own locally-resolved values, not attacker-supplied input - same reasoning already accepted for skills/semgrep. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> --------- Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com> Co-authored-by: toolhive-release-app[bot] <280093410+toolhive-release-app[bot]@users.noreply.github.com> Co-authored-by: Dan Barr <danbarr@users.noreply.github.com> Co-authored-by: Dan Barr <6922515+danbarr@users.noreply.github.com> Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
skill-security-scan.zeroize-audit: allowlistedBEHAVIOR_EVAL_SUBPROCESS— verified against the actual upstream source, both flaggedsubprocess.run()calls are list-form (noshell=True) with fixed commands/scripts and no attacker-controlled arguments.Not fixed here
The other 5 skills bumped in this digest (
agentic-actions-auditor,codeql,constant-time-analysis,sharp-edges,yara-rule-authoring) still failskill-security-scan. Their logs showFailed to parse meta-analysis response: No valid JSON found in response, the known scanner meta-analyzer bug — dozens to hundreds of raw pattern-match findings survive as blocking when this happens, which isn't practically fixable by hand-allowlisting. This is a scanner tooling reliability issue, not a dockyard content problem.Test plan
dockhandlocally, ranvalidate-skillagainstskills/zeroize-audit/spec.yaml—Status: VALIDzeroize-auditto go green, the other 5 to remain red pending the scanner bugCo-Authored-By: Claude Sonnet 5 noreply@anthropic.com