The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing external software dependencies (claude-codemem via NPM and a Homebrew cask from the MadAppGang tap) if they are not already present on the system.
[REMOTE_CODE_EXECUTION]: The skill executes the claudemem CLI tool and various shell utilities (e.g., find, stat, grep, wc) via the Bash tool to perform structural analysis of the codebase.
[PROMPT_INJECTION]: The instructions use forceful directives such as "STOP," "MANDATORY," and "FORBIDDEN" to override standard agent capabilities (like grep or find) and enforce the use of the claudemem tool. Additionally, it establishes a surface for indirect prompt injection by processing external codebase data.
Ingestion points: Data returned from claudemem --raw commands.
Boundary markers: Absent; there are no instructions to ignore embedded directives in the ingested code analysis.
Capability inventory: Uses Bash, Task, and Read tools across the workflow.
Sanitization: None detected; the skill does not explicitly sanitize or escape the data received from the external tool before processing.
[COMMAND_EXECUTION]: The skill uses Bash to check versioning, verify index freshness, and perform complex AST queries using the external claudemem tool.

ultrathink-detective