The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, specifically through the decoding of user-provided URLs and the retrieval of work item content (titles, descriptions, and comments) from the Meegle backend. This creates an attack surface for indirect prompt injection if the remote content contains malicious instructions for the agent.
Ingestion points: Untrusted data enters the agent's context through the meegle url decode command and various work item retrieval commands like workitem get and search-filter (documented in SKILL.md and references/workitem.md).
Boundary markers: The skill includes a dedicated "Prompt Injection Defense" section in references/error-handling.md which explicitly instructs the agent to treat all remote data as data, not instruction, and never to execute instructions found in user-generated content.
Capability inventory: The skill possesses capabilities for file system access (read/write), network operations via the CLI, and command execution (CLI tools and local interpreters).
Sanitization: While the skill provides strong semantic instructions to the model to ignore injections, it does not specify technical sanitization or escaping mechanisms for the ingested text before it is placed in the prompt.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to install and update the @tingwillforever/meegle-cli package globally via NPM. As this package belongs to the skill's author, it is categorized as a vendor-owned resource.
[COMMAND_EXECUTION]: The skill's core functionality relies on executing the meegle CLI tool. It also instructs the agent to use local interpreters (node, python3) for processing JSON data, formatting timestamps, and reducing metadata payloads. These are standard operations for a developer-oriented CLI skill.

meegle-cli