The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Twitter/X, including user tweets, search results, and notifications.
Ingestion points: Untrusted data enters the agent context via opencli twitter search, timeline, and notifications commands as documented in SKILL.md and references/commands.md.
Boundary markers: The instructions do not define clear delimiters or specify "ignore instructions" warnings to prevent the agent from obeying malicious prompts embedded in the retrieved tweet content.
Capability inventory: The skill executes shell commands using the opencli tool and can write output to local files (CSV, JSON, YAML).
Sanitization: There is no evidence of sanitization or filtering of the external tweet content before it is processed by the agent.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains a shell command that executes automatically when the skill is loaded by compatible platforms.
Evidence: !(command -v opencli && opencli doctor 2>&1 | head -5 && echo "READY" || echo "SETUP_NEEDED") 2>/dev/null || echo "NOT_INSTALLED"inSKILL.md`.
Analysis: This command performs a diagnostic check on the opencli tool's installation and health. While functional and diagnostic in this case, it utilizes a mechanism for silent shell command invocation at load time.
[EXTERNAL_DOWNLOADS]: The skill relies on external software components and instructions for the user to install third-party dependencies.
Evidence: The documentation instructs the user to run npm install -g @jackwener/opencli and to manually download and install a Chrome extension from https://github.com/jackwener/opencli/releases.
[COMMAND_EXECUTION]: The skill's primary functionality relies on executing the opencli binary with various flags and subcommands to fetch web data, which represents direct shell command execution.

twitter-reader