The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the claude-repo-mem Python package from a public registry and downloads an embedding model (BAAI/bge-small-en-v1.5) from a well-known machine learning repository during the indexing process.
[COMMAND_EXECUTION]: Executes several subcommands of the claude-repo-mem utility, including index to build the repository map, serve to launch the MCP server, and distill to process transcripts. It also modifies the repository configuration by optionally installing a git post-commit hook.
[DATA_EXFILTRATION]: The skill is designed to read and process the entire target repository, including source code and documentation, to build a local vector index. It also accesses Claude Code session transcripts stored in ~/.claude/projects/ to extract and persist durable project decisions.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted data from the repository and interaction logs.
Ingestion points: The indexing process (claude-repo-mem index) reads every file within the target repository workspace, and the distill command reads external transcript files.
Boundary markers: No specific delimiters or "ignore instructions" markers are defined for the data retrieved from the index.
Capability inventory: The skill exposes 11 tools (e.g., recall, trace, expand) that return indexed content directly to the agent's context.
Sanitization: No explicit sanitization or filtering of the content retrieved from the repository or transcripts is performed before it is presented to the model.

memory-management