sdlc-close-task

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of shell commands involving git, grep, and python3 to interact with project repositories. These commands interpolate user-provided inputs such as the task key, code folder path, and commit hashes, which could lead to command injection if inputs are not properly sanitized by the AI agent.
  • Evidence found in SKILL.md and references/write-protocol.md showing interpolation of <Task key>, <code folder>, and <commit> into shell command strings.
  • Example: git -C "$SRC" log "$BASE" --grep='^Task: <Task key>$' --format='%H %cI %s'
  • Example: git -C "$SRC" worktree remove "<the worktree>".
  • [DYNAMIC_EXECUTION]: The skill is configured to execute a Python script located within the local project's directory structure.
  • Evidence in references/read-protocol.md defines the tool as T="python3 .wiki-llm/tools/wiki_llm.py". This tool is central to the skill's drafting and validation logic.
  • [INDIRECT_PROMPT_INJECTION]: The skill reads and processes Markdown files from the wiki repository, which serves as an ingestion surface for potentially malicious content that could influence agent behavior.
  • Ingestion points: The skill reads Task files, Design files, parent overviews, and log files (SKILL.md Section 1, read-protocol.md Section 3).
  • Boundary markers: There are no specific delimiters or instructions to treat ingested file content as untrusted data.
  • Capability inventory: The skill possesses the ability to execute shell commands, perform Git network and write operations, and run local scripts.
  • Sanitization: No sanitization or validation of the content of the ingested files is described before it is used to determine subsequent actions.
  • [EXTERNAL_DOWNLOADS]: The skill performs network operations to keep local repositories synchronized with their remotes.
  • Evidence in SKILL.md and read-protocol.md shows the use of git fetch --quiet origin to update both the code and wiki repositories.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 09:20 AM