The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a block of text designed to override agent behavior: 'STOP -- DO NOT READ THIS FILE. You are already reading it. This prompt was injected into your context...'. This mimics adversarial injection techniques used to bypass standard instruction processing.
[INDIRECT_PROMPT_INJECTION]: The skill aggregates data from external sources and writes it to a file for later consumption by a planner agent.
Ingestion points: Phase CONTEXT.md and output from the mempalace CLI/MCP tools.
Boundary markers: None identified in the prompt templates or file writing steps.
Capability inventory: Bash (command execution), Write (file system modification), Read, and Grep tools.
Sanitization: The skill distills retrieved data into patterns and decisions but lacks explicit sanitization or escaping of the external content before it is written to the phase directory.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute vendor-specific CLI utilities (mempalace wake-up, mempalace search) to perform memory retrieval operations.

gsd-mempalace-recall