The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from a 'ledger' containing project summaries, task lists, and journals. This external content is untrusted and could be used to inject malicious instructions into the agent's context.
Ingestion points: Data returned by the delphi resume command as described in SKILL.md.
Boundary markers: No boundary markers or 'ignore' instructions are present to isolate the ingested data from the agent's operational logic.
Capability inventory: The agent has capabilities to perform file system reads, execute git commands, and run a verify script.
Sanitization: The skill provides no instructions for sanitizing or validating the content retrieved from the project ledger.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands that incorporate external and potentially unsafe inputs.
The command delphi resume $ARGUMENTS --json directly uses user-supplied arguments, which could lead to command injection if the input is not sanitized by the platform.
The agent is directed to 'run the verify command where it is cheap,' executing a script defined within the potentially untrusted project ledger, posing a risk of arbitrary code execution.

delphi-resume