The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches an external specification from https://agentskills.io/specification.md and explicitly instructs the agent to "read the returned file completely" to guide the skill creation process. It also retrieves documentation from code.claude.com for the same purpose. This allows external content to influence the agent's behavior during the scaffolding process.
Ingestion points: scripts/fetch-agentskills-spec.sh (downloads specification.md to a local cache) and WebFetch (retrieves documentation from code.claude.com).
Boundary markers: The skill does not provide clear delimiters or instructions to the agent to ignore potentially malicious content within the fetched markdown files.
Capability inventory: The agent is authorized to create directories (mkdir), write files (SKILL.md, openai.yaml), create symlinks (ln -s), change file permissions (chmod 755), and run validation tools (ai-skillet).
Sanitization: No validation or filtering is performed on the fetched content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill uses a bundled script (scripts/fetch-agentskills-spec.sh) to download content from https://agentskills.io/specification.md. The script uses curl with secure flags (--proto '=https', --fail) and manages a local cache with restrictive permissions (umask 077).
[COMMAND_EXECUTION]: The scaffolding workflow involves executing several shell commands, including mkdir, ln -s, chmod 755, and the tool ai-skillet doctor. These commands are necessary for the skill's primary purpose of project initialization but represent a broad set of capabilities.
[PRIVILEGE_ESCALATION]: The skill instructs the agent to perform chmod 755 on all scripts and tests in the newly scaffolded project. While typical for generating executable scripts, this involves programmatically granting execution permissions to files created during the workflow.

skill-writing