pre-flight-check

Pass

Audited by Gen Agent Trust Hub on Oct 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill surfaces content from the .learnings/, .evals/, and .context-surfing/ directories, which acts as a vector for potentially malicious instructions from prior sessions to influence the current agent state.\n
  • Ingestion points: The skill reads from LEARNINGS.md, ERRORS.md, HEALS.md, EVAL_INDEX.md, and handoff files located in .context-surfing/.\n
  • Boundary markers: While the scripts/pre-flight.sh output is wrapped in <pre-flight-check> XML-style tags, the manual instructions in SKILL.md lack explicit delimiters or warnings for the agent to treat the accumulated knowledge as untrusted data.\n
  • Capability inventory: The skill uses a bash script to perform file scans and instructs the agent to manually read the contents of the identified directories.\n
  • Sanitization: There is no evidence of sanitization or instruction-filtering applied to the markdown content before it is surfaced to the agent.\n- [COMMAND_EXECUTION]: The skill executes a local shell script (scripts/pre-flight.sh) which uses standard system utilities such as grep, awk, and find. The script uses hardcoded directory paths and does not process external or user-supplied input as command arguments, minimizing the likelihood of command injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 27, 2026, 09:21 PM