The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to perform repository audits and version checks.
Evidence: It reads plugin cache files (cat ~/.claude/plugins/cache/...), uses the GitHub CLI (gh release view), and queries the version of the local bridge utility (node ~/.iskron-bridge/iskron-bridge.mjs --version).
[PERSISTENCE]: The skill modifies local configuration files to install lifecycle hooks and plugins that run across different sessions.
Evidence: It updates .claude/settings.json with SessionStart, PostToolUse, and PreToolUse hooks. It also deploys JavaScript-based plugins into the .opencode/plugins/ directory to manage session rituals.
[DYNAMIC_EXECUTION]: The skill uses dynamic code loading to initialize specialized sub-agents.
Evidence: Sub-agents are configured to load a vendor-provided bridge script from the user's home directory (~/.iskron-bridge/iskron-bridge.mjs) using the node -e command.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository data to generate and update agent instructions in files like AGENTS.md.
Ingestion points: Reads metadata and configuration from package.json, README.md, Makefile, and various CI/workflow YAML files.
Boundary markers: Uses distinct YAML frontmatter blocks and specific 'source' fields (e.g., 'Граф — источник': 'выведено') to distinguish extracted facts from manual directives.
Capability inventory: The skill has extensive file-writing permissions, subprocess execution capabilities, and the ability to spawn sub-agents with varying model tiers.
Sanitization: It includes a mandatory 'Verification' step (Step 7) where it cross-references generated output against the source files to ensure accuracy and prevent instructions from diverging from repository facts.

iskronify