The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands to orchestrate external processes using cmux and tmux as described in references/transports.md.
[PROMPT_INJECTION]: The skill instructs the agent to use the --trust-all-tools flag when launching sub-agents in references/transports.md. This instruction explicitly disables the safety guardrails that require user confirmation for tool execution, creating a security bypass for the delegated agent.
[DATA_EXFILTRATION]: The skill reads repository content to create a task brief. This pattern of reading potentially sensitive local data and transmitting it to an external agent provider via the sub-agent constitutes a data exposure risk.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection attacks because it ingests untrusted data and uses it to drive sub-agent behavior.
Ingestion points: User input from {{ARGUMENTS}} and project file content from .delegate-brief.md are passed directly to the sub-agent.
Boundary markers: There are no clear delimiters or instructions within the brief to separate user-provided task descriptions from system instructions.
Capability inventory: The skill facilitates the execution of arbitrary terminal commands and file system operations via the delegated sessions managed by tmux and cmux.
Sanitization: The skill lacks any evidence of input validation, escaping, or filtering for data before it is processed by the sub-agent.

afk-delegate