The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a dynamic path resolution mechanism using Bash shell expansion: ${CLAUDE_PLUGIN_ROOT:-$(ls -td "$HOME"/.claude/plugins/cache/insight-wave/cogni-portfolio/*/ | head -1)}. This logic is executed whenever the agent calls internal utility scripts to resolve the correct plugin directory at runtime.
[REMOTE_CODE_EXECUTION]: The skill invokes several external Bash scripts located in the plugin's directory, including project-status.sh, validate-entities.sh, cascade-rename.sh, and append-claim.sh. These scripts are executed via the Bash tool to perform project management, entity validation, and data persistence tasks.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting and processing data from various JSON files and external research results. This content is interpolated into prompts for both the main agent and multiple specialized subagents.
Ingestion points: Market definitions (markets/), customer profiles (customers/), competitive intelligence (competitors/), and context files (context/), plus external web research data.
Boundary markers: None explicitly defined in the instructions to separate untrusted data from agent instructions.
Capability inventory: Includes file modification (Write, Edit), shell command execution (Bash), and task delegation to other agents (Agent).
Sanitization: No specific sanitization or escaping of interpolated data is described.
[EXTERNAL_DOWNLOADS]: The skill utilizes a subagent to perform web research for industry benchmarks, competitor claims, and evidence enrichment. This involves the retrieval and processing of data from external, uncontrolled internet sources.

propositions