domains / general / analyze / analyze

analyze

Skill card

combined

The two skills share the name 'analyze' but perform completely different tasks. The first is a security skill for analyzing blast radius and attack paths in AI infrastructure using the 'agent-bom' CLI. The second is a research skill for end-to-end empirical data analysis (R/Stata/Python) that dispatches specialized sub-agents (Coder, Data-engineer, coder-critic). There is no overlap in procedure, tools, or domain, so crowning one would hide a real alternative.

Repositories

Other copies

Combined skill

# Analyze

This skill name covers two distinct workflows. Determine the user's intent to select the correct procedure.

## 1. Security: Blast Radius & Attack Path Analysis
**Use when:** User asks about "blast radius", "threat intel", "risk score", "attack path", or "lateral movement" in AI infrastructure.

**Procedure:**
1. Ensure `agent-bom` is installed (`pipx install agent-bom`).
2. Run `agent-bom agents --verbose` to map blast radius details.
3. Run `agent-bom graph` to generate the agent context graph.
4. Use `blast_radius(cve_id="...")` to map specific CVE impacts.
5. Present findings clearly. If CRITICAL CVEs are found, ask the user if they want a remediation plan.

**Guardrails:**
- Analysis is read-only.
- Only public CVE IDs are sent externally.
- No internal data leaves the machine.

## 2. Research: End-to-End Data Analysis
**Use when:** User provides a dataset or analysis goal for empirical research (R, Stata, Python, Julia).

**Procedure:**
1. **Context Gathering:** Read `domain-profile.md`, strategy memos, and `CLAUDE.md` for language preferences.
2. **Data Preparation:** If raw data is provided, dispatch **Data-engineer** to clean data, handle missing values, and create descriptive figures.
3. **Main Analysis:** Dispatch **Coder** to run the main specification, robustness checks, and generate publication-ready tables/figures. Must produce `results_summary.md`.
4. **Code Review:** Dispatch **coder-critic** to run a 12-category checklist (strategy alignment, sanity checks, reproducibility, etc.).
5. **Fix Issues:** If critical issues are found, re-dispatch Coder (max 3 rounds).
6. **Present Results:** Summarize estimates, scripts, outputs, and review scores.

**Key Artifacts:**
- `results_summary.md` (Mandatory)
- Tables in `paper/tables/`
- Figures in `paper/figures/`
- Code review report in `quality_reports/`