v0.68.0 - Wider workflow-audit inputs
ContextForge v0.68.0
This release expands contextforge workflow-audit so agentic GitHub workflow checks cover attacker-controlled titles and branch/ref fields, not only Markdown bodies or comments.
Highlights
- Treats issue titles, pull request titles, PR head refs/labels,
github.head_ref,github.ref_name, review-comment bodies, and discussion titles as untrusted agent inputs. - Adds regression coverage for title/ref text flowing into a Claude-powered workflow with write permissions.
- Refreshes README, workflow-audit docs, LLM discovery files, generated artifacts, changelog, and research notes.
- Links the research note to GitHub Actions script-injection guidance, GitHub Security Lab untrusted-input guidance, and recent agentic workflow injection research.
Verification
- PR #161 passed CI and ContextForge Audit on GitHub.
- Main commit
6315b88passed CI and ContextForge Audit. - Local validation included red/green regression testing, typecheck, full tests, build, security benchmark, audit, workflow-audit artifact generation, publish-readiness, and npm pack dry-run.