The Agent Skills Directory

[SAFE]: The skill is a utility for organizing and searching local markdown files used as agent memory. It uses standard shell commands to perform file operations within a defined workspace.
[COMMAND_EXECUTION]: The provided scripts (init.sh, detect.sh, organize.sh, search.sh, categorize.sh, snapshot.sh, stats.sh) execute common Unix utilities such as mkdir, mv, cp, grep, find, and jq to manage memory files. All paths are relative to the user's home directory or workspace, and no dangerous system-level commands are used.
[DATA_EXFILTRATION]: Analysis of the shell scripts confirms no network activity or external data transmission. The skill does not use curl, wget, or any other networking tools to communicate with remote servers.
[INDIRECT_PROMPT_INJECTION]: The skill processes historical interaction data stored in memory files. While this creates a surface for indirect prompt injection if the agent previously recorded untrusted content, the functionality is core to the skill's purpose as a memory manager.
Ingestion points: Reads markdown files from episodic, semantic, and procedural subdirectories in ~/.openclaw/workspace/memory/.
Boundary markers: Content from memory files is aggregated into snapshots or search results without explicit delimiters or instructions to the agent to treat the content as data rather than instructions.
Capability inventory: The skill allows the agent to move, copy, and read files within the workspace using the provided scripts.
Sanitization: The scripts do not perform validation or sanitization of the file content before displaying it. Additionally, the categorize.sh script lacks sanitization for the target filename, which could allow for path traversal within the local file system if the agent is directed to use malicious paths.

memory-manager