aw-create-plan

Pass

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local shell utilities within SKILL.md to manage its internal state.\n
  • It executes git branch --show-current to determine the working directory path.\n
  • It uses mkdir, ls, sed, sort, and tail to calculate the next version number for plan snapshots.\n- [PROMPT_INJECTION]: The skill in SKILL.md provides an attack surface for indirect prompt injection by persisting chat context into permanent project files.\n
  • Ingestion points: Data from the current conversation (which may include untrusted user input or external data) is ingested during Step 2 to populate the plan template.\n
  • Boundary markers: The plan template lacks delimiters or explicit instructions to ignore potentially malicious commands embedded in the processed text.\n
  • Capability inventory: The skill possesses the ability to write files (plan.md, plan.vN.md) and execute shell commands. The generated plan is specifically intended to be used by future AI sessions, which typically have broader system access.\n
  • Sanitization: No sanitization, escaping, or validation logic is applied to the content before it is written to the file system.
Audit Metadata
Risk Level
SAFE
Analyzed
May 20, 2026, 11:46 AM
Security Audit — agent-trust-hub — aw-create-plan