The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Node.js scripts and system commands to validate, install, and activate configuration files.
Evidence: SKILL.md invokes node scripts/install-mode.mjs and node scripts/configure-telegram.mjs to perform filesystem mutations.
Evidence: SKILL.md uses npx claude-mem restart and npx claude-mem status to restart the local worker service.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a factory for system prompts, creating a surface where malicious or poorly-formed user input could be persisted into the agent's long-term memory configuration.
Ingestion points: SKILL.md uses interactive questions (AskUserQuestion) to gather domain-specific data from the user.
Boundary markers: Absent; the skill encourages the agent to 'infer' prompts and recording policies directly from the user's work descriptions.
Capability inventory: The skill can write files to the modes/ directory, modify the global settings.json, and restart the claude-mem worker (documented in SKILL.md and scripts/install-mode.mjs).
Sanitization: scripts/install-mode.mjs performs strict JSON schema validation and ID pattern checks but does not sanitize or filter the natural language content within the generated prompts.
[DATA_EXFILTRATION]: The skill facilitates the transmission of observation metadata to the Telegram Bot API as a core feature.
Evidence: scripts/configure-telegram.mjs uses the fetch API to communicate with https://api.telegram.org for bot authentication and message delivery.
Note: This targets a well-known service and is the intended functionality of the skill; the skill also implements security best practices by prompting for tokens via hidden terminal input and setting 0600 file permissions on configuration backups.

mode-creator