# Import

> ---

- Skill: `together2329/import` (Agent Skill)
- Install (CLI): `npx skillmds@latest add together2329/import`
- Raw SKILL.md: https://api.skillmd.com/api/skills/together2329/import/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: together2329 (https://skillmd.com/u/together2329)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/together2329/import

---

---

## name: import

description: Import requirement documents, notes, existing RTL, or legacy YAML into ssot-gen evidence before grill-me/to-ssot. Use when the user wants to build SSOT from IP docs or existing RTL without hand-authoring the SSOT directly.

# Import Evidence Into SSOT-Gen

This skill imports evidence only. The output of this step is an SSOT evidence
summary and QA backlog; the production YAML is still written later by
`/to-ssot`.

## Rules

1. **SSOT remains the authority.** Imported RTL, docs, spreadsheets, or legacy
  YAML are evidence, not downstream artifacts to preserve blindly.
2. **Do not write production RTL/TB/sim files.** This workflow may write only
  SSOT-side notes, import manifests, QA records, and `<ip>/yaml/<ip>.ssot.yaml`
   when explicitly converting with `/to-ssot`.
3. **Separate known facts from guesses.** Facts found in imported material can
  be used directly. Conflicts or missing behavioral facts must become QA cards.
4. **Existing RTL is structural evidence.** Extract module names, ports,
  parameters, reset/clock intent, register decode hints, FSM states, memories,
   and obvious protocol names. Do not infer unimplemented behavior from signal
   names alone.
5. **Use auto-select only when enabled.** In `auto-select` mode, `ask_user`
  chooses suggested/default answers and records approved QA. In normal
   interactive mode, user answers are required.

## Process

1. Resolve the IP name from command args, active session, path names, or an
  existing `<ip>/` directory. Ask once if ambiguous.
2. Read each imported path:
  - Markdown/text/requirements: use `read_file`; for non-text docs, use
   `read_doc` when available.
  - RTL/SystemVerilog/Verilog: read top-level files and filelists first; use
  `grep_file`/`find_files` for modules, ports, parameters, always blocks,
  state encodings, and register address constants.
  - YAML/JSON: parse as legacy SSOT/evidence when possible.
3. Create or update `<ip>/req/import_manifest.json` with:
  - imported path list
  - detected evidence type (`doc`, `rtl`, `legacy_yaml`, `filelist`, `other`)
  - extracted candidate facts
  - conflicts and missing facts
  - recommended next command (`/grill-me <ip>` or `/to-ssot <ip>`)
4. Create or update `<ip>/req/extracted_decisions.json` with the same
  per-decision facts and source excerpts in a compact machine-readable shape.
5. Record executable handoff work in `workflow_todos.<stage>[]`. Each item must
  include `command`, `script`, `instructions`, `content`, `detail`, `criteria`,
   and `source_refs`; `/to-ssot` consumes these as its template-backed todo
   ledger.
6. Record non-blocking missing decisions with `record_ssot_qa`. Use `ask_user`
  only when the import cannot determine the IP name, top module, or basic IP
   type needed for the next write.
7. End with `[SSOT IMPORT]`:
  - IP
  - evidence paths read
  - facts imported
  - conflicts / pending QA
  - next step

## RTL Evidence Extraction Checklist

- `module` declarations and top-module candidate
- parameter names/defaults
- clock/reset ports and reset polarity
- interface ports grouped by prefix/protocol
- register address constants and bitfields
- FSM state names and transition conditions
- memory/FIFO instances or arrays
- interrupt/error/debug/status signals
- comments that describe requirements

Do not treat a stub, tieoff, or placeholder module as behavioral proof.