The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes production run history via the terse history command and project source code. This data constitutes a potential surface for indirect prompt injection, as malicious content embedded in historical event payloads or logs could attempt to influence the agent's behavior during refactoring.
Ingestion points: Reads project files in src/terse.jobs.ts, src/jobs/, and production run history (including chat history and trigger payloads) via the terse CLI.
Boundary markers: The instructions lack explicit delimiters for separating ingested history data from current instructions, though they do require explicit narration and user confirmation for high-impact actions.
Capability inventory: The skill possesses significant capabilities, including modifying source code, deploying workflows via terse deploy, replaying production events, and utilizing computer-use tools for UI automation.
Sanitization: There is no evidence of automated sanitization or filtering for incoming historical data payloads.
[COMMAND_EXECUTION]: The skill utilizes several CLI tools to perform its tasks, including terse (for history, replay, testing, and deployment), tsc (for typechecking), and base64 (for encoding secrets). These commands are standard for the development and maintenance of Terse workflows.
[DATA_EXFILTRATION]: The skill accesses official documentation and backend services at docs.useterse.ai. These are recognized as legitimate vendor resources for the 'terseai' platform.

terse-improve