The Agent Skills Directory

[COMMAND_EXECUTION]: The bootstrap script auth.sh uses sudo to create a symlink in the system-wide /usr/local/bin directory, which represents a request for elevated administrative privileges on the host system.
[COMMAND_EXECUTION]: The skill executes a bundled binary (bin/tg) and a shell script (auth.sh) distributed within the skill's file structure.
[DATA_EXFILTRATION]: The skill instructions direct the agent to read the user's SSH public key from the sensitive file path ~/.ssh/id_ed25519.pub and transmit it to the Tangled service during registration.
[CREDENTIALS_UNSAFE]: The skill retrieves and processes sensitive ATProto/Bluesky app passwords from environment variables (ATPROTO_APP_PASSWORD and BSKY_APP_PASSWORD). While standard for this protocol, these secrets are handled by the local bootstrap script.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by processing untrusted content from external Tangled issues and pull requests.
Ingestion points: Data fetched via tg issue view and tg pr view from external repositories.
Boundary markers: No specific delimiters or safety instructions are provided to the agent to distinguish between platform data and user instructions.
Capability inventory: Execution of local scripts/binaries, network communication, and repository cloning into the local workspace.
Sanitization: The skill lacks explicit sanitization or validation mechanisms for content retrieved from external sources before it is processed by the agent.

tg