modelscope · diff
git:20260622.3e8b757 to git:20260902.0c46844
8 added, 74 removed. Audit A to C.
+ <!-- Generated Claude Code shim from .agents/skills/modelscope/SKILL.md. Do not edit. -->
---
name: modelscope
description: "Download, resume, status-check, and SHA256-verify ModelScope model weights. Use for $modelscope download/status/verify/check, Chinese requests to 下载/续传/补全/查看进度/校验 ModelScope 权重, and tasks that need durable background ModelScope downloads under explicit local directories."
---
# ModelScope
- Use the bundled scripts from this skill directory. Prefer the compact manager first:
-
- - `scripts/modelscope_auto.py` - status, auto-resume, background download, and post-download verification
- - `scripts/download_from_modelscope.py` - low-level single-model downloader
- - `scripts/modelscope_download_status.py` - low-level size status
- - `scripts/verify_modelscope_sha256.py` - low-level SHA256 verification
-
- Do not inline long `nohup`/`setsid` shell blocks. Do not read or tail large logs unless a task fails or the user asks.
-
- ## Model Mapping
-
- Represent every model as `MODEL_ID=LOCAL_DIR`.
-
- - `MODEL_ID` must be `namespace/name`.
- - `LOCAL_DIR` must be explicit.
- - If the user says “to `/root`” without a model subdirectory, use `/root/namespace/name`.
- - Use revision `master` unless specified.
- - Repeat `--model MODEL_ID=LOCAL_DIR` for multiple models.
-
- ## Download / Resume / Auto Complete
-
- For `$modelscope download`, resume, repair-after-approval, or “check and continue if incomplete”, run:
-
- ```bash
- python3 "$SKILL_DIR/scripts/modelscope_auto.py" ensure \
- --model "$MODEL_ID=$LOCAL_DIR" \
- --revision "$REVISION"
- ```
-
- `ensure` behavior:
-
- - If a task is active, leave it running and report compact status.
- - If official files are incomplete and no task is active, start a detached background worker in the same `LOCAL_DIR`.
- - If files are complete but verification is missing or stale, start detached SHA256 verification.
- - If verification reports real missing, size mismatch, or SHA256 mismatch, report it and ask before repair.
- - It preserves partial files and never deletes weights.
-
- The manager writes `download.pid`, `download.launch.log`, `download.log`, `verify.log`, `modelscope_sha256.report.json`, `modelscope_sha256.tsv`, and `SHA256SUMS` in `LOCAL_DIR`.
-
- Proxy options:
-
- - Pass no proxy option by default.
- - Add `--no-proxy` only when requested.
- - Add `--proxy "$PROXY_URL"` only when provided.
-
- ## Status
-
- For explicit status only:
-
- ```bash
- python3 "$SKILL_DIR/scripts/modelscope_auto.py" status \
- --model "$MODEL_ID=$LOCAL_DIR" \
- --revision "$REVISION"
- ```
-
- If the user asks for all tasks, require a root and run `--root ROOT`; the script discovers `*/download.pid` and infers `namespace/name` from the last two path components.
-
- Report only the compact script output: state, percent, local/expected size, PID, verification state, and directory. Include log paths only when useful.
-
- ## Verify
-
- For explicit verification:
-
- ```bash
- python3 "$SKILL_DIR/scripts/modelscope_auto.py" verify \
- --model "$MODEL_ID=$LOCAL_DIR" \
- --revision "$REVISION"
- ```
+ Canonical skill source:
- Verification ignores `.gitattributes` by default because it is Git metadata, not model weight content. Do not repair or redownload solely because `.gitattributes` is absent.
+ `.agents/skills/modelscope/SKILL.md`
- ## Output Rules
+ Before using this skill:
- - Keep responses short.
- - Do not paste large command output or progress bars.
- - Summarize each model as `state`, percent, PID, verification result, and paths.
- - If network or filesystem sandboxing blocks a required command, rerun with approval as needed.
+ 1. Read the canonical skill file above.
+ 2. Follow its routing rules, entrypoints, guardrails, and acceptance criteria.
+ 3. Use `.remote-dev` companion tools for ordinary remote endpoint read/edit/bash/search/patch work.
+ 4. Use this Claude project skill only for the domain workflow described by the canonical source.