The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from historical session transcripts and journals (.memsearch/memory/*.md). If these records contain malicious instructions from previous sessions, they could influence the agent during the distillation process.
Ingestion points: Reads journal files in .memsearch/memory/ and transcript files via memsearch transcript <file>.
Boundary markers: No specific delimiters are defined for the content being read, though instructions advise the agent to be exact and look for recurring procedures.
Capability inventory: Uses the Bash tool to execute memsearch commands and write files.
Sanitization: Instructions explicitly warn to exclude secrets and verify details against original transcripts.
[DYNAMIC_EXECUTION]: The skill dynamically generates new executable content (AI agent skills) by drafting markdown steps and persisting them to the file system.
Evidence: The skill uses printf and memsearch skills add to create new SKILL.md files based on interpreted session history.
[COMMAND_EXECUTION]: The skill relies heavily on the Bash tool to interact with the local memsearch CLI, managing configuration, listing candidates, and installing files to sensitive directories like .claude/skills and ~/.claude/skills.

memory-to-skill