The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses highly authoritative language under the 'IRON LAW' section, employing directives like 'NEVER use', 'ALWAYS use', and 'not negotiable' to force the agent to bypass its standard tool selection logic in favor of a specific local binary.
[COMMAND_EXECUTION]: The skill's primary operations rely on executing a local binary file located at ~/projects/consensus-cli/consensus. It also uses shell pipes to link curl and python execution.
[INDIRECT_PROMPT_INJECTION]: The skill defines a process for resolving DOIs by interpolating them into a curl command string. Because the DOI is sourced from external search results and then placed directly into a shell command without explicit instruction for escaping or sanitization, it creates a potential surface for command injection if a malicious search result is processed.
Ingestion points: Academic paper DOIs retrieved from the consensus CLI tool.
Boundary markers: Absent. The DOI is interpolated directly into a shell command string.
Capability inventory: The skill uses the consensus binary, curl, and python3 via uv run in SKILL.md.
Sanitization: None specified for the interpolation of the DOI variable into the shell command.
[DYNAMIC_EXECUTION]: The skill utilizes uv run python3 -c to execute a multi-line Python script provided as a string to parse JSON output from a network request at runtime.
[EXTERNAL_DOWNLOADS]: The skill fetches metadata from CrossRef (api.crossref.org), which is a well-known academic service used for DOI resolution.

consensus