The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The workflow in SKILL.md instructs the agent to execute cargo run or cargo install --path . when the kt CLI is not found. This leads to the execution of arbitrary code from the repository's local environment.
[COMMAND_EXECUTION]: The kt land and kt undo commands described in references/cli-behavior.md perform high-impact history-rewriting operations, specifically git push --force-with-lease and git push --force, which can cause data loss.
[CREDENTIALS_UNSAFE]: The documentation in references/cli-behavior.md explicitly lists environment variable names used for storing API keys, such as KITE_OPENAI_API_KEY and OPENAI_API_KEY, facilitating potential discovery of secrets.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection during the kt land process.
Ingestion points: Git status and diff content are processed by the skill (referenced in references/cli-behavior.md).
Boundary markers: None identified; the diff is passed directly to LLM providers.
Capability inventory: The skill can execute subprocesses, rewrite Git history, and perform network requests (referenced in SKILL.md and references/cli-behavior.md).
Sanitization: No evidence of sanitization or escaping of the diff content before LLM processing was found.
[EXTERNAL_DOWNLOADS]: The skill connects to local and remote API endpoints (http://localhost:11434/api/chat and OpenAI-compatible URLs) to generate content via external LLM providers.

use-kite