The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill maintains a persistent memory system by logging user corrections and self-reflections, which are then used to influence future agent responses. This creates a surface where external input can modify the agent's operational logic.
Ingestion points: Untrusted user input is captured via corrections and logged into files such as corrections.md and memory.md within the ~/Clawic/data/self-improving/ directory.
Boundary markers: The skill includes a boundaries.md file that specifies types of sensitive information (e.g., credentials, financial data) that must not be stored, and provides rules for when to ask for user consent.
Capability inventory: The agent is instructed to modify workspace configuration files (AGENTS.md, SOUL.md, HEARTBEAT.md) and perform file operations (write, append, read) across its memory tiers.
Sanitization: The skill relies on instructions and policy-based guidelines to prevent the storage of sensitive information, rather than programmatic filtering of input content.
[COMMAND_EXECUTION]: The initialization and maintenance processes use shell commands to manage the environment and list files.
Evidence: setup.md contains mkdir and touch commands for directory initialization, and shell loops to list files in the domains and projects directories.
[REMOTE_CODE_EXECUTION]: The setup instructions facilitate the installation of companion software from the vendor's ecosystem.
Evidence: setup.md includes the command npx clawic add proactivity, which downloads and executes the proactivity skill. This tool is provided by the skill's vendor.
[EXTERNAL_DOWNLOADS]: The skill uses package managers to fetch additional components.
Evidence: The npx clawic add command initiates a download from the registry to add functionality.
[DYNAMIC_EXECUTION]: The skill generates and runs shell commands dynamically as part of its memory management and setup flow.
Evidence: In setup.md, a bash loop is used to find and list markdown files in the memory directories, which is then interpreted by the agent.
[PERSISTENCE]: The skill establishes persistent behavioral changes by updating the agent's core steering files.
Evidence: Instructions in setup.md guide the agent to append specific logic to AGENTS.md, SOUL.md, and HEARTBEAT.md, establishing a recurring loop for learning and maintenance across sessions.

self-improving