wiki-ingest

Pass

Audited by Gen Agent Trust Hub on May 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes local Python helper scripts using subprocess.run in scripts/rebuild_context_brief.py and scripts/convert_to_md.py. These calls use the list-based argument format and avoid shell-specific features.\n- [COMMAND_EXECUTION]: The scripts/convert_to_md.py script shells out to an optional external utility, markitdown, to handle conversion of binary office formats (e.g., .docx, .pptx) to markdown.\n- [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection due to the processing of untrusted data:\n
  • Ingestion points: Processes files from the raw/ directory, content fetched from arbitrary URLs, and user-provided text pastes (documented in SKILL.md Steps 2 and 2a).\n
  • Boundary markers: While it uses templates with placeholders (defined in references/templates/), there are no strict adversarial boundary markers to prevent malicious instructions in the source content from influencing the agent's behavior during page generation.\n
  • Capability inventory: Possesses capabilities to write to the file system (via atomic_write in scripts/_wikilib.py), update configuration files (scripts/config_bump.py), and execute local scripts.\n
  • Sanitization: Implements regex-based slug sanitization (SLUG_CLEAN_RE in scripts/batch_builder.py), but lacks robust filtering for instructions embedded within the source documents themselves.
Audit Metadata
Risk Level
SAFE
Analyzed
May 19, 2026, 04:41 PM
Security Audit — agent-trust-hub — wiki-ingest