The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation instructions require cloning a repository from an external, unverified GitHub source (github.com/MercuriusDream/agent-estate.git) rather than using the author's verified infrastructure.
[REMOTE_CODE_EXECUTION]: The skill requires the user to grant execution permissions (chmod +x) to shell scripts (setup-estate.sh, stop-hook.sh) which are then executed within the local environment to manage the agent's lifecycle.
[COMMAND_EXECUTION]: A shell hook (stop-hook.sh) is used to intercept the agent's 'Stop' event. The script programmatically blocks the exit command by returning a 'block' decision and re-injecting prompts, overriding standard process termination behavior.
[PROMPT_INJECTION]: The skill contains explicit instructions to override the agent's normal operational boundaries, commanding it to "never stop" and "immediately start the next cycle" regardless of task completion.
[INDIRECT_PROMPT_INJECTION]: The skill implements a ledger system (.claude/agent-estate.md) that acts as a cross-context memory bridge.
Ingestion points: The agent is instructed to read the ledger file at the beginning of every cycle (SKILL.md).
Boundary markers: There are no sanitization boundaries or "ignore embedded instruction" warnings present in the ledger handling logic.
Capability inventory: The skill possesses the capability to modify local files, execute shell scripts, and control process exit status.
Sanitization: No sanitization is performed on the "Tell The Next Claude" section of the ledger, allowing instructions from one execution cycle to influence the behavior of the next cycle.

agent-estate