The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to install components from external marketplaces and repositories, including alirezarezvani/claude-skills on a marketplace and self-improving-agent via clawhub.
[COMMAND_EXECUTION]: Installation for the Codex CLI platform involves executing a local shell script ./scripts/codex-install.sh, which performs automated setup tasks.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection as it ingests and processes untrusted data from multiple sources.
Ingestion points: The memory-analyst and error-capture components read MEMORY.md, project-specific topic files, and Bash command error outputs.
Boundary markers: There are no documented delimiters or boundary markers to prevent the agent from interpreting instructions embedded within the memory data or error logs.
Capability inventory: The skill has the capability to write to primary configuration files (CLAUDE.md, .claude/rules/*.md) and generate new SKILL.md files through the skill-extractor agent.
Sanitization: The analysis does not indicate any sanitization or validation of patterns extracted from memory before they are promoted to durable project rules.

self-improving-agent