scrolls-compact

Pass

Audited by Gen Agent Trust Hub on Nov 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on local git command execution to function.
  • The core logic in scripts/compact.py uses subprocess.run to execute commands like git blame, git rev-parse, git show, git commit, and git tag.
  • These operations are used to identify the age of content and manage the checkpointing process. Commands are constructed using lists (avoiding shell interpolation) and target local repository data.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes content from documentation files that may contain untrusted data.
  • Ingestion points: Reads files like SPEC.md, PLAN.md, and HANDOFF.md from the project directory (specified in SKILL.md).
  • Boundary markers: Absent. The skill reads raw Markdown content without delimiters separating data from instructions.
  • Capability inventory: The skill has the capability to write to the local filesystem and perform git commits/tags.
  • Sanitization: The --finalize mode performs validation to ensure only designated "old" units were modified, providing a safeguard against unintended edits by the agent.
  • [EXTERNAL_DOWNLOADS]: The skill references an external resource for its maintenance components.
  • The maintenance script scripts/build_example_viewer.py and the resulting assets/example-viewer.html reference the marked library from cdnjs.cloudflare.com.
  • This resource is fetched from a well-known service and is used only for rendering documentation examples in a browser.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 18, 2026, 01:27 AM