The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external feature requests and local wiki content to generate technical specifications. This creates a surface where instructions embedded in the processed data could influence the agent's behavior during the drafting process.
Ingestion points: User-supplied feature or change requests; local wiki content from concepts/, entities/, and sources/adr/ (SKILL.md).
Boundary markers: The skill does not define specific delimiters or instructions for the agent to ignore potentially malicious embedded directives in the source data.
Capability inventory: Writing files to the local filesystem (llmwiki/wiki/sources/draft/, llmwiki/html/), executing local Python scripts (code-logger.py, unknown-ledger.py), and running shell commands via Node.js (archify.mjs) (SKILL.md).
Sanitization: No sanitization or validation of the input request or wiki content is mentioned before interpolation into the draft spec.
[DYNAMIC_EXECUTION]: The skill dynamically generates architectural specifications in JSON format based on the proposed plan and then executes a local Node.js utility to process these files into HTML diagrams.
Evidence: The skill generates *.sequence.json files and executes node ~/.agents/skills/archify/bin/archify.mjs deliver sequence ... to render them (SKILL.md).
[COMMAND_EXECUTION]: The skill executes local scripts and binaries to manage task logging and architectural rendering.
Evidence: The skill instructions include the execution of python3 harness/scripts/code-logger.py, python3 harness/scripts/unknown-ledger.py, and the invocation of the archify tool via node (SKILL.md).

propose