The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads and writes to local markdown files in the .agents/ directory to maintain context about the user and their team. This is a standard practice for agent memory and does not involve access to sensitive system files, environment variables, or hardcoded credentials.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from locally stored profile files. While this creates a theoretical surface for malicious instructions if the profile files are externally tampered with, the skill does not possess dangerous capabilities like network requests or shell execution to facilitate an exploit.
Ingestion points: .agents/em-context.md, .agents/reports/[name].md
Boundary markers: None explicitly defined.
Capability inventory: Markdown file read/write operations; no shell access, network operations, or dynamic code execution.
Sanitization: No explicit sanitization or validation of the profile content.
[EXTERNAL_DOWNLOADS]: Reference links to educational content on the author's newsletter (manager.dev) and Substack are provided. These links are for informational purposes only and do not trigger automated downloads, package installations, or remote code execution.

managing-urgency