Skip to content
Back to search

x-saves-harvest

SkillWorks with Claude CodeGitHub

Turn the user's X (Twitter) bookmarks and likes into a backlog of Claude skills, MCP servers, repos and techniques in the Obsidian vault. Use when asked to check, harvest or go through X saves, bookmarks or likes, "what did I save on X for Claude", "x-saves", or for the weekly X saves run. Read-only on X.

Safety scan

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

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/chrstphrpond--x-saves-harvest/badge.svg)](/listings/chrstphrpond--x-saves-harvest)

HTML

<a href="/listings/chrstphrpond--x-saves-harvest"><img src="/api/listings/chrstphrpond--x-saves-harvest/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

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.

SKILL.md · Manifest · 99 lines
  1. ---
  2. name: x-saves-harvest
  3. description: Turn the user's X (Twitter) bookmarks and likes into a backlog of Claude skills, MCP servers, repos and techniques in the Obsidian vault. Use when asked to check, harvest or go through X saves, bookmarks or likes, "what did I save on X for Claude", "x-saves", or for the weekly X saves run. Read-only on X.
  4. ---
  5. # X saves harvest
  6. Collects posts from X through an Orca tab, keeps the new AI-tool ones, expands them with full text and repo facts, and appends what is worth keeping to `Notes/X Saves for Claude.md` in the notes vault. Remembers handled posts, so each run shows only new saves.
  7. `S` below is this skill's folder; `RUN` is `~/.local/state/x-saves-harvest/runs/<YYYY-MM-DD>`.
  8. ## 1. Preflight
  9. - `orca status` shows `runtimeReachable: true`. If not, ask the user to open Orca.
  10. - Use `--worktree path:<this session's working directory>` written out literally. The `orca-worktree-guard` hook rejects variables and other worktrees.
  11. - `python3 S/scripts/xsaves.py status` shows how many posts are handled and the last runs.
  12. - `mkdir -p RUN`.
  13. ## 2. Collect
  14. Each command opens its own Orca tab, scrolls in small steps, saves the posts and closes the tab:
  15. ```
  16. python3 S/scripts/xsaves.py collect --source bookmarks --worktree path:/abs/worktree --expect-account <your handle> --out RUN/bookmarks.json
  17. python3 S/scripts/xsaves.py collect --source likes --worktree path:/abs/worktree --expect-account <your handle> --out RUN/likes.json
  18. ```
  19. - It stops at the end of the list, or after 15 handled posts in a row. A first run on a long likes list takes 15–20 minutes, so run it in the background.
  20. - The feed is optional sampling: `--source feed --max-steps 60`.
  21. - A topic the user asks for: `--source search --query "claude blender" --max-steps 40 --stop-after-known 0 --out RUN/search-<topic>.json` (X's Top tab). Search posts are not saves, so put them in their own `### From search: <query>` group at the end of the run section.
  22. - If the user opens something else in the collector's tab, the script stops with "tab moved to …" and keeps what it had. Rerun that source.
  23. - Wrong account: the script stops. Ask the user to switch accounts in the Orca tab, then rerun.
  24. - Few posts and "end of list" too early means X stopped loading because the tab is not drawn. Rerun with `--focus`, which brings the tab forward in Orca, and tell the user why.
  25. ## 3. Expand
  26. ```
  27. python3 S/scripts/xsaves.py expand RUN/bookmarks.json RUN/likes.json --out RUN/digest.json --md RUN/digest.md
  28. ```
  29. It drops handled posts, keeps the AI-related ones by keyword, fetches full text, X articles and quoted posts from the public fxtwitter API, and checks every linked GitHub repo with `gh` (stars, last push, `SKILL.md` folders inside). If it reports 0 new posts, say so and stop.
  30. ## 4. Judge
  31. Read all of `RUN/digest.md` (in chunks if long). Give each post one kind:
  32. | Kind | Keep when |
  33. |---|---|
  34. | Skill | the repo contains `SKILL.md` folders, or the post gives a `skills add` command |
  35. | MCP server | it adds tools to Claude through MCP |
  36. | Repo or tool | open-source project or app that works with Claude |
  37. | Technique | a reusable prompt, workflow or method, with enough detail in the post to use |
  38. | Resource | free API, asset or footage library, prompt gallery, inspiration site |
  39. | Build idea | something worth building as our own skill |
  40. | Drop | hype with nothing usable; "comment to get it" with no method in the post; off-topic; or unsafe (cracking paid software, impersonating real people, spam) |
  41. - Merge posts about the same repo or tool into one entry and link the most useful post.
  42. - "same name as an installed skill" is only a name match. Confirm with `orca skills installed` before writing "Installed".
  43. - Note overlaps with installed skills, for example HyperFrames, media-use or impeccable.
  44. ## 5. Write the vault note
  45. File: `~/notes/X Saves for Claude.md` (set this to your own notes folder). Create it from the template below if missing. Otherwise insert the new run section directly under the intro, newest first, with a targeted edit. Never rewrite earlier runs.
  46. ```
  47. # X Saves for Claude
  48. #reference #x-saves #claude-code
  49. Claude skills, MCP servers, repos and techniques from my X bookmarks and likes, newest run first. Written by the `x-saves-harvest` skill; handled post ids are in `~/.local/state/x-saves-harvest/seen.json`.
  50. ## 2026-10-25
  51. 41 new saves checked, 23 kept.
  52. ### Skills
  53. - [emilkowalski/skills](https://github.com/emilkowalski/skills): animation review, vocabulary and design-engineering skills. 49k stars, pushed 2026-10-11. [post](https://x.com/emilkowalski/status/2077404975555031509)
  54. ```
  55. - Groups in this order, empty ones left out: Skills, MCP servers, Repos and tools, Techniques, Resources, Build ideas.
  56. - One line per entry: link, what it does in 20 words or fewer, facts (stars and last push for repos), then `[post](url)`. Add "Installed." or "Overlaps <skill>." when true. For techniques, quote the key instruction in under 40 words.
  57. - Vault rules: no emojis, no editorializing, and no `- [ ]` task lines, so the task list stays clean.
  58. ## 6. Mark handled and report
  59. Only after the note is saved:
  60. ```
  61. python3 S/scripts/xsaves.py mark-seen RUN/digest.json --note "kept 23"
  62. ```
  63. Then report in chat: posts collected, new, AI-related, kept and dropped; the three to five best new finds, one line each; and the note path.
  64. ## Rules
  65. - Read-only on X: never like, bookmark, follow, reply or post, and never sign in for the user.
  66. - Only public post ids go to fxtwitter; repo names go to the GitHub API. Nothing else leaves the machine.
  67. - Suggest installs; do not install anything found here without the user's go-ahead.
README.md · README · 33 lines
  1. # x-saves-harvest
  2. A Claude Code skill I run once a week. It goes through my X bookmarks and likes, keeps the new posts about AI tools, checks each linked GitHub repo, and adds what is worth keeping to a note. It only reads; it never likes, follows, replies or posts.
  3. Written by Christopher Pond Maquidato, October 2026.
  4. ## What it does
  5. 1. **Collect.** Opens my bookmarks and likes in a browser tab I am already signed in to, scrolls in small steps and saves the posts. It stops at the end of the list or after 15 posts in a row it has already handled.
  6. 2. **Expand.** Drops handled posts, keeps the AI-related ones, fetches the full text, and checks every linked GitHub repo with `gh` (stars, last push, whether it holds skills).
  7. 3. **Judge.** Claude sorts each post into a kind (skill, MCP server, repo, technique, resource, build idea) or drops it.
  8. 4. **Write.** Adds one dated section to the note, newest first. Earlier runs are never rewritten.
  9. 5. **Mark handled.** Only after the note is saved, so a failed run loses nothing.
  10. ## What keeps it reliable
  11. - **It checks the account first.** `--expect-account` stops the run if the browser is signed in to a different account.
  12. - **It remembers what it handled.** Post ids are kept in `~/.local/state/x-saves-harvest/seen.json`, so each run shows only new saves and a rerun never doubles an entry.
  13. - **It stops loudly.** If the tab is moved, or the list stops loading early, the script says so and keeps what it had.
  14. - **It never signs in for me.** If the session has expired it asks me to sign in.
  15. ## What's here
  16. | File | What it is |
  17. |---|---|
  18. | `SKILL.md` | What Claude reads: the steps, the sorting rules, the note format |
  19. | `scripts/xsaves.py` | Collecting, expanding and the handled-posts ledger. Plain Python, no packages |
  20. ## Needs
  21. - Claude Code, with this folder copied to `~/.claude/skills/x-saves-harvest/`.
  22. - The Orca app, whose `orca` CLI drives the browser tab the collector reads, and the `gh` CLI for repo facts.
  23. - Your own X handle for `--expect-account`, and your own notes path in step 5 of `SKILL.md`.

Ratings and comments

Sign in to rate this skill.

Sign in

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