The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes session history and transcripts to identify patterns, which constitutes a significant attack surface for indirect prompt injection.
Ingestion points: Session history and transcript data are ingested via joelclaw sessions signals, joelclaw sessions search, and joelclaw session inspect.
Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions within the processed transcript data.
Capability inventory: The skill has the capability to execute shell commands (find, rg, xargs, joelclaw) and can propose or create new skill assets (Pass 2).
Sanitization: While the skill instructs the agent not to include raw transcript dumps or capture secrets, there is no technical sanitization or validation of the ingested session content described.
[COMMAND_EXECUTION]: The skill uses shell commands that incorporate user-provided or session-derived strings, which could lead to command injection if not properly handled by the agent.
Evidence: Commands such as joelclaw sessions search "<candidate query>" and rg -n "...|<candidate-term>" interpolate variables directly into the command line.
Tool Usage: The skill utilizes the joelclaw CLI tool, which appears to be a local vendor resource associated with the author, to perform session analysis and asset management.

signal-friction-scan