The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill grants the agent permission to perform sensitive repository actions autonomously, including merging pull requests and force-pushing branches with --force-with-lease. This bypasses standard human-in-the-loop review processes for code changes.
[PROMPT_INJECTION]: The instructions contain explicit overrides for safety and workflow constraints, stating that a 'full send' request 'replaces' any existing rules in project files or other instructions that require the agent to ask before testing or leave testing to the user.
[INDIRECT_PROMPT_INJECTION]: The skill enables an autonomous loop where the agent processes external inputs such as CI logs and test results to 'babysit' and fix changes.
Ingestion points: External data sources include CI failure logs, test output, and repository-level instruction files.
Boundary markers: No specific delimiters or 'ignore embedded instruction' markers are defined for the data ingested during the autonomous loop.
Capability inventory: The agent has the power to execute shell commands (tests), force-push code, and merge pull requests.
Sanitization: There is no evidence of sanitization or validation of the external content before it is used to influence the agent's next steps.
[COMMAND_EXECUTION]: The agent is authorized to use broad testing methods including browser automation, simulators, and dev servers without seeking prior approval, provided they serve as 'direct proof' for the task.

full-send