← ClaudeAtlas

threat-modellisted

Build or regenerate the PASTA threat model for the current repository. Runs the threat-modeler subagent through the PASTA phases (S1 Objectives → S2 Scope → S3 Decomposition → S4 Threats) and writes .kuzushi/threat-model.json plus an ASCII data-flow diagram.
allsmog/kuzushi-security-plugin · ★ 0 · AI & Automation · score 74
Install: claude install-skill allsmog/kuzushi-security-plugin
# Regenerate the PASTA threat model Build (or regenerate, overwriting any existing one) the PASTA threat model for this repository. 1. Run the prepare step to open a run and get the stage-file paths + scope inputs: `node "${CLAUDE_PLUGIN_ROOT}/scripts/cmd/threat-model-prepare.mjs" --target "<repo root>"` (use the project's working directory as the target). 2. Work the PASTA phases in order — S1 Objectives → S2 Scope → S3 Decomposition → S4 Threats — writing each `pasta-s*.json` in the schema your agent definition specifies. Gather evidence with `tree_sitter:*` (AST, taint sources/sinks, callers) and ambient LSP while reading files. Use `codeql:query` / `joern:query` only if a database/CPG already exists. **Do NOT run semgrep** — SAST scanning is out of scope for threat modeling. 3. Run the assemble step it prints to persist `.kuzushi/threat-model.json` and render the ASCII data-flow diagram. 4. Report a brief summary (threat counts by category + top threats), then paste the ASCII data-flow diagram from `.kuzushi/threat-model-dfd.txt` verbatim **inside a triple-backtick fenced code block** (```). Mandatory — it's column-aligned ASCII and breaks if pasted as prose. Paste the file contents as-is; do not re-draw or summarize the diagram. ## When NOT to use - To find or confirm concrete vulnerabilities — the model names *threats* to investigate; `/threat-hunt`, `/taint-analysis`, and `/systems-hunt` do the finding. - To run SAST — semgrep is exp