The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx to download and execute the @catalyst-cloud/catalyst-skills package if the CLI tool is not detected on the local system. This ensures the agent uses the vendor-specific tools required for the tenant's workflows.
[COMMAND_EXECUTION]: All primary operations are performed by spawning the catalyst-skills CLI tool via node:child_process.spawnSync. The scripts use array-based argument passing to mitigate shell injection risks and strictly follow the permitted tools defined in the skill metadata.
[DATA_EXFILTRATION]: The scripts read a local configuration file at ~/.config/catalyst-cloud/customer.json to retrieve API keys or OAuth tokens. This access is limited to obtaining credentials necessary for authenticating with the Catalyst Cloud API for intended skill operations.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading human-generated comments from Linear tickets to record decisions. The settle.mjs script extracts and interpolates this text into release notes for other tickets, which presents an ingestion surface for potentially malicious content.
Ingestion points: Ticket comment bodies are read in settle.mjs to extract answer details.
Boundary markers: Absent; the extracted text is interpolated directly into a summary string without specific delimiters.
Capability inventory: The skill has the capability to execute specific CLI commands (cli.mjs) and write comments or change ticket states via the vendor platform (settle.mjs, raise.mjs).
Sanitization: The untrusted input is truncated to a maximum of 200 characters and whitespace is trimmed in settle.mjs before being used in downstream operations.

what-needs-me