The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation and instructions reference the execution of local scripts for data processing and visualization, such as scripts/ingest.py, scripts/server.py, and node scripts in the skills/ directory. These commands are intended to be executed by the user or an agent to enhance the skill's analytical capabilities within its specific research domain.
[EXTERNAL_DOWNLOADS]: The README and installation guides provide commands to download the skill and its source repositories using git clone and npx skills add. These are standard installation procedures for the platforms the skill targets (Claude Code, Cursor, Codex) and point to the project's own vendor-controlled resources.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, such as social media posts, news, and regulatory filings, presenting a surface for indirect prompt injection.
Ingestion points: Described in SKILL.md (Step 5 and Step 9), where the agent reads data from X (Twitter), industry reports, and SEC/CSRC filings.
Boundary markers: The skill enforces a strict evidence-tiering system (Confirmed, Management Claim, My Inference, Pure Speculation) to categorize and delimit external information from analytical instructions.
Capability inventory: The skill uses subprocess calls (node, python3) to run internal scripts and accesses local file-system databases (SQLite) for data storage, as noted in the 'Data Enhancement' sections.
Sanitization: The workflow includes mandatory 'Independent Review' (Step 7) and 'Data Discipline' rules (Step 9), specifically a 'Fabrication Guard' and 'Freshness Guard', to validate the integrity and source of external data before integration into reports.

serenity-unified-skill