The Agent Skills Directory

[DYNAMIC_EXECUTION]: Several modules (manifest.sh, uninstall-memo-flow.sh, user-registry.sh) use string interpolation to generate Python code dynamically for JSON processing. Shell variables containing project paths, manifest data, and identifiers are inserted directly into python3 -c command strings using single or triple quotes. This creates a risk of Python code injection if any of these values contain characters that break out of the string delimiters.
[INDIRECT_PROMPT_INJECTION]: The uninstallation logic processes a manifest.json file from the project root. Since this file resides within the project being audited/processed, a malicious repository could include a crafted manifest designed to exploit the unsafe Python interpolation patterns to execute arbitrary code during the uninstallation process.
Ingestion points: .claude/memo-flow/manifest.json file content and project directory paths.
Boundary markers: No delimiters or validation logic exists to prevent embedded instructions or code from being interpreted during processing.
Capability inventory: The skill can delete files (rm), write to files via Python, and modify the global user registry located at ~/.claude/memo-flow/registry.json.
Sanitization: The scripts do not consistently use safe argument passing methods like sys.argv and instead rely on vulnerable shell-to-python string concatenation.
[EXTERNAL_DOWNLOADS]: The SKILL.md file references an installation command using npx skills@latest add GuillermoMurillo/memo-flow. This points to the author's official repository and is used for setup instructions.
[COMMAND_EXECUTION]: The main uninstall script executes various file system operations and internal helper scripts based on the keys and values parsed from the project's manifest file.

uninstall-memo-flow