The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of Bash tools for file manipulation, including rm for file deletion and python3 for running a validation script. While it instructs the agent to limit these operations to specific memory directories, the underlying tools have broad capabilities.
[PROMPT_INJECTION]: The skill is designed to read and process all markdown files in the harness and .memory/ directories, which may contain content from external sources (e.g., cloned repositories). This creates a surface for indirect prompt injection where malicious instructions embedded in memory files could influence the agent's behavior during consolidation.
Ingestion points: The skill reads every *.md file in the harness and .memory/ directories using Read and Grep tools.
Boundary markers: The skill does not specify the use of delimiters or 'ignore instructions' warnings when interpolating memory content during the consolidation process.
Capability inventory: The skill possesses file-writing (Write, Edit), file-deletion (rm), and shell execution (Bash, python3) capabilities.
Sanitization: Validation is performed via an internal script (validate-consolidate.py) and a second-pass sub-agent review, though specific data sanitization for file content is not detailed.

consolidate