Skip to content
Back to search

research-box

MarketplaceWorks with Claude CodeGitHub

Paolo Pignatelli's research-box Agent Skill, packaged as a Claude Code plugin.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 20 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/PaoloTCS/research-box with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/PaoloTCS/research-box'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/paolotcs--research-box/badge.svg)](/listings/paolotcs--research-box)

HTML

<a href="/listings/paolotcs--research-box"><img src="/api/listings/paolotcs--research-box/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

The skill follows the open Agent Skills format (agentskills.io). Copy the `research-box` folder into your agent's skills directory, or upload it as a skill in an app that supports skills. You can also install it with the skills CLI:

npx skills add PaoloTCS/research-box

In Claude Code, the repository is also a plugin marketplace, so the skill installs as a plugin in one step:

/plugin install research-box --marketplace PaoloTCS/research-box

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 15 lines
  1. {
  2. "name": "research-box",
  3. "owner": {
  4. "name": "Paolo Pignatelli",
  5. "url": "https://github.com/PaoloTCS"
  6. },
  7. "description": "Paolo Pignatelli's research-box Agent Skill, packaged as a Claude Code plugin.",
  8. "plugins": [
  9. {
  10. "name": "research-box",
  11. "source": "./",
  12. "description": "Bundle related research work into named boxes with sharing levels that only narrow, RO-Crate metadata and checksummed snapshot drops."
  13. }
  14. ]
  15. }
README.md · README · 103 lines
  1. # research-box
  2. An Agent Skill for keeping related research work together: documents, spreadsheets, diagrams, code, data, chats and agent tasks bundled into a named **box**, with sharing levels that only narrow.
  3. Say "make a box called prime-numbers", "add this sheet to the box", "drop the box about prime numbers", "open box prime-numbers" or "give the box to my other agent", and the agent knows what to do.
  4. - **Live layer:** links that stay current. Agents of the same assistant hand each other links, not files.
  5. - **Snapshot layer:** a dropped folder with exported files, a readable `BOX.md`, an RO-Crate 1.1 `ro-crate-metadata.json`, and SHA-256 checksums of every file.
  6. - **Sharing levels:** private, same assistant, shareable. A drop for wider readers leaves out what is not cleared for them, writes a clean copy that holds nothing else, and says what it left out.
  7. - **Checks:** `--check` tells you whether any file in a drop, including its own description, changed or was copied stale, and a list of watch words flags private terms that are about to leave in a shareable drop.
  8. ## Quick demo
  9. From a clone of this repository, using the example box in `examples/prime-numbers`:
  10. ![A terminal session: a shareable drop of the example box keeps two items, --check passes, then fails after one file is edited.](docs/quick-demo.gif)
  11. ```
  12. $ python3 research-box/scripts/build_box.py examples/prime-numbers --for shareable --out /tmp/prime-share
  13. prime-numbers: 2 item(s) kept, 2 file(s) checksummed, written to /tmp/prime-share
  14. Left out for shareable readers: Working notes, Planning chat
  15. Watch words in code/simulate.py: Project Lantern (advisory; --strict refuses the drop)
  16. $ python3 research-box/scripts/build_box.py /tmp/prime-share --check
  17. All files match the manifest.
  18. $ printf '\n# edited\n' >> /tmp/prime-share/sheet/results.csv
  19. $ python3 research-box/scripts/build_box.py /tmp/prime-share --check
  20. Changed since the last drop: sheet/results.csv
  21. ```
  22. `--check` exits 0 when the drop matches its manifests and 1 after the edit. Add `--strict` to the first command and the drop is refused until the watch word is removed.
  23. ## Contents
  24. ```
  25. research-box/
  26. SKILL.md instructions the agent follows
  27. references/box-format.md the box.json format and how it relates to RO-Crate and BagIt
  28. scripts/build_box.py writes BOX.md, RO-Crate metadata and checksum manifests; checks drops
  29. assets/box.example.json an example box
  30. LICENSE Apache License 2.0
  31. NOTICE copyright and attribution
  32. .claude-plugin/ plugin manifest, marketplace file and icon
  33. examples/prime-numbers/ a small box to try the script on
  34. tests/test_build_box.py unittest suite, standard library only
  35. ```
  36. `build_box.py` uses the Python standard library only.
  37. ## Try it
  38. The commands and the output in [Quick demo](#quick-demo) were run against this revision. To run the 23-test suite:
  39. ```
  40. python3 -m unittest discover tests
  41. ```
  42. The example box has one item at each sharing level and a watch word left in a shareable file, so each safeguard shows. The tests cover the drop rules, labels, path containment, watch words and the check's exit status.
  43. Manifests can also be verified without the script: `shasum -a 256 -c manifest-sha256.txt`.
  44. ## Install
  45. The skill follows the open Agent Skills format (agentskills.io). Copy the `research-box` folder into your agent's skills directory, or upload it as a skill in an app that supports skills. You can also install it with the skills CLI:
  46. ```
  47. npx skills add PaoloTCS/research-box
  48. ```
  49. In Claude Code, the repository is also a plugin marketplace, so the skill installs as a plugin in one step:
  50. ```
  51. /plugin install research-box --marketplace PaoloTCS/research-box
  52. ```
  53. ## What it runs
  54. The skill tells the agent to run `research-box/scripts/build_box.py`, a Python script that uses the standard library only. It reads and writes files in the box folder you name and in the `--out` folder, and nothing else. It makes no network calls, sends nothing anywhere, and never deletes files. The live links a box lists are opened by the agent with its own tools, not by the script.
  55. ## Versions
  56. - **0.2.5** (November 19, 2026), from the first real user's bug report: watch words match whole words and phrases, not substrings, so a short private term no longer matches inside ordinary words such as "file" and blocks every strict drop; each hit is reported with its file, line and context; held-back state lines and box details are reported like held-back items. 27 tests.
  57. - **0.2.4** (November 19, 2026): packaged as a Claude Code plugin (`.claude-plugin/plugin.json` and a one-plugin marketplace), with an icon and a note on what the script runs. No change to the skill or the script.
  58. - **0.2.3** (November 19, 2026), after a third outside review: state-line labels survive repeated exports; fields outside the format are never copied into a drop; a missing declared snapshot, or one file listed under two sharing levels, refuses the drop; `--check` returns 2 (incomplete) without a tag manifest; watch words are scanned in every generated file, and `--strict` refuses a shareable drop that contains them; clearer wording on what checksums do and do not prove. 23 tests.
  59. - **0.2.2** (November 19, 2026), after a second outside review: snapshot paths must stay inside the box folder (absolute paths, `..` and outside symlinks are refused before anything is written, and `--check` flags manifest paths outside the folder); `--help` and usage errors print usage instead of a traceback; a runnable example box and a 15-test suite (27 tests as of 0.2.5).
  60. - **0.2.1** (November 19, 2026), after review by an early user: `--out` is required for any drop that holds something back, so held-back files can never be left beside shared ones; `--check` exits 1 on any changed, missing or unlisted file and 2 with no manifest, so pipelines can gate on it.
  61. - **0.2** (November 19, 2026), after the first real drops: box details and state lines follow the box's sharing level, with an optional `public` block for wider drops; `--out` writes a clean filtered copy; `--check` compares a drop with its manifest and a BagIt-style tag manifest, after every copy and before a refresh; watch words (`private_terms`); item revisions; registry label mapping (`level_labels`).
  62. - **0.1** (November 19, 2026): first version.
  63. ## Dedication
  64. Released November 19, 2026, with thanks to the people at Anthropic who made Claude, which helped build it.
  65. ## Credits
  66. Created by **Paolo Pignatelli**. Developed in collaboration with **Claude**, an AI model made by **Anthropic**, which drafted and tested the skill text, the manifest format and the build script (November 2026).
  67. This project is not affiliated with or endorsed by Anthropic.
  68. ## License
  69. Apache License 2.0. See `LICENSE` and `NOTICE`; copies sit inside `research-box/` so the skill folder is complete on its own. If you redistribute the skill, keep the NOTICE file, which carries the attribution above.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.