The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts bin/fm-update.sh and bin/fm-send.sh to perform system updates and send notifications.
[REMOTE_CODE_EXECUTION]: The skill performs a Git fast-forward from a remote source (origin). This process downloads and applies updates to the agent's executable scripts (bin/) and logic, effectively allowing remote code to modify the agent's runtime environment.
[PROMPT_INJECTION]: This skill exhibits an indirect prompt injection surface by instructing the agent to re-read AGENTS.md (and the symlinked CLAUDE.md) immediately after pulling remote changes. This allows the remote Git repository to override or alter the agent's core operating instructions.
Ingestion points: Remote Git repository (origin) data is merged into local files, specifically AGENTS.md.
Boundary markers: None identified; the agent is instructed to read the file directly to "refresh your operating instructions."
Capability inventory: The skill utilizes bin/fm-update.sh (shell script execution) and bin/fm-send.sh (inter-process communication/messaging).
Sanitization: The skill relies on Git's "fast-forward only" logic to prevent disruptive merges, but it does not validate the content of the instructions being pulled.
[DYNAMIC_EXECUTION]: The skill facilitates self-modifying behavior by updating and then reloading its own instruction set (AGENTS.md) and binaries at runtime.

updatefirstmate