The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from an "inbox" and "observations" which may contain malicious instructions designed to influence the agent's actions (Indirect Prompt Injection).
Ingestion points: Data enters the context via kspec inbox list (referenced in SKILL.md and docs/inbox.md) and kspec meta observations (referenced in docs/observations.md).
Boundary markers: While explicit text-based delimiters are not used, the "Record → Act" pattern serves as a functional boundary by separating data ingestion from command execution.
Capability inventory: The agent can execute impactful actions through kspec triage act and kspec batch across all provided files.
Sanitization: The workflow relies on the agent's evaluation and the recording of reasoning before execution to ensure intended behavior.
[COMMAND_EXECUTION]: The skill uses the kspec CLI tool to perform all operations, which is the intended functional method for this skill.
Evidence: All primary workflow steps in SKILL.md and documentation files use tool calls like kspec triage record, kspec task set, and kspec batch to manage state and execute tasks.

kspec-triage