The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform system maintenance. Specific commands include pnpm code:index:reindex for search indexing, pnpm validate:full for framework validation, and node .claude/lib/memory/memory-rotator.cjs for memory management. While these are typical for environment maintenance, they represent a powerful capability that is triggered automatically by the scheduler.
[PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection. Its primary purpose is to automate the reading and processing of external data on a recurring basis.
Ingestion points: Files such as issues.md, CHANGELOG.md, and external data sources like Telegram messages are read into the agent's context during scheduled loops.
Boundary markers: The provided task prompts (e.g., "Read issues.md and CHANGELOG.md and give morning briefing") do not include explicit delimiters or instructions to ignore embedded commands within the ingested data.
Capability inventory: The skill has access to Bash, CronCreate, Read, and TaskUpdate, which could be abused if the agent follows malicious instructions found within the processed files.
Sanitization: There is no evidence of sanitization or validation of the content read from these external files before it is processed by the model.
[DATA_EXFILTRATION]: While the skill contains examples of interacting with external services (e.g., "Check for new Telegram messages"), these are presented as user-controlled examples. However, the automated nature of these tasks could be leveraged for data exfiltration if the agent's instructions are compromised via the indirect injection surface described above.

scheduled-tasks