agentleFS
bendruckerbendrucker/claude/.claude/skills/agent-ideas/SKILL.md

agent-ideas

Skill18 starsChanged 27 days agoMITView on GitHub

Harvest agent-tooling ideas from prominent developers.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Write the digest to tmp/agent-ideas-digest-<YYYY-Www>.md:

    line 53
  2. Distinctive:

    Keep the digest in this session's context as well.

    line 59
  3. Very specific:

    Do not tag the doorway item claude-code.

    line 79

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Agent Ideas

Harvest agent-tooling ideas from a curated list of thinkers and map them to concrete artifacts in this repo (a skill, hook, setting, or rule). The harvest runs in two phases: a feed pass that fetches, mines, and delivers a digest, and a local pass that fills the X-only gap and files the keepers. As the weekly routine these run headless-then-local, bridged by a teleport doorway item. Run on-demand at your own terminal, they collapse into one pass with no doorway (see Deliver the Doorway Item).

Sources

packages/agent-ideas/sources.ts is the prunable core: a version-controlled list of thinkers, each with a feed URL, or marked x-only when no usable feed exists. Edit it directly to add, drop, or retune sources. Keep it small and high-signal; a source earns its place by regularly producing ideas that map to artifacts in this repo.

Fetch

Pull the last 8 days of posts across all feed sources:

bun ${CLAUDE_SKILL_DIR}/../../../packages/agent-ideas/fetch.ts > tmp/agent-ideas-posts.json

Each result is {source, sourceType, feedUrl, posts:[{title,url,date,excerpt}], error?}. Sources with error set could not be fetched; note them but don't block. x-only sources are absent here by design (the local phase handles them after teleport).

Narrow to specific sources or widen the window with flags:

bun ${CLAUDE_SKILL_DIR}/../../../packages/agent-ideas/fetch.ts --days 14 --source simon --source mario

Mine

Turn posts into repo-mappable idea cards. The bar is high: a card must name a real surface in this repo it would refine, or make a clear case for a net-new artifact. See references/mining.md for the card schema, the proven heuristics (per-source cap by depth, auto-demote 404 sources, dedupe against existing repo surfaces), and the fan-out flow.

Fan out one mining Agent call per source (batch thin sources together). Give each the source's posts, the schema, the heuristics, and a read-only view of the repo so it can fill repoOverlap and dedupe. Collect the cards, verify each sourceUrl resolves (demote 404s), then dedupe across agents.

Synthesize

Write the digest to tmp/agent-ideas-digest-<YYYY-Www>.md:

  • A top-line summary: idea count, the standout cards, the low/medium/high split.
  • Per-idea cards in full (the schema fields), grouped high → low actionability.
  • A short "repo overlaps" note listing cards that refine existing surfaces.

Keep the digest in this session's context as well. The local phase reads it from here.

Deliver the Doorway Item

Skip this section on a local on-demand run. The doorway exists only to bridge the headless weekly routine to a local terminal; claude --teleport ${CLAUDE_SESSION_ID} minted from a session you are already in points at itself. When the browser and Things are reachable (you invoked the skill yourself), go straight from Synthesize to Fill the X-Only Gap. The doorway is the remote routine's handoff, nothing more.

For the weekly routine, create exactly one Things inbox item. It is the routine's only notification, so make its notes self-sufficient:

  • Title: [agent-ideas] Weekly digest: N ideas (YYYY-Www)
  • Notes: the top-line summary, then the teleport command on its own line so it's tappable:

    claude --teleport ${CLAUDE_SESSION_ID}

Pick the delivery path by environment:

  • Local (Things reachable on this Mac): use the things:url skill's inbox capture. Leave it untagged, since the local phase adds claude-code per keeper later.
  • Remote (headless routine, no local MCPs): use the Zapier connector tool mcp__claude_ai_Zapier__things_create_to-do. It reaches Things through Anthropic's proxy. Tags aren't available remotely, which is fine: the doorway item is intentionally untagged.

Do not tag the doorway item claude-code. Tagging happens per keeper in the local phase, so the whole digest doesn't land in the improve-claude-code backlog as one blob.

Local Phase

The local phase fills the X-only gap and files the keepers. You reach it two ways. The weekly routine gets here when you tap the doorway item: claude --teleport pulls the cloud session into a local terminal where the browser and Things are reachable. A local on-demand run flows straight here from Synthesize, with no doorway in between.

The digest is already in this session's context. After a teleport into a fresh terminal it may not be, so read the latest tmp/agent-ideas-digest-*.md to recover the feed-derived cards.

Fill the X-Only Gap

x-only sources were skipped in the remote fetch. Read their recent timelines in the browser and mine them into cards with the same bar as the feed path. See references/browser.md for the tool-loading sequence, the per-author flow, and the failure modes (extension not connected, logged-out wall). If the browser is unavailable, note which authors went unread and proceed with the feed cards.

Capture Keepers

For each keeper card, create one Things todo via the things:url skill's inbox capture, tagged claude-code so improve-claude-code picks it up:

  • Title: [agent-idea] <card title>
  • Notes: the pitch, then the source URL, then the suggested artifactType. This is the shape improve-claude-code consumes (short title, full notes).
  • Pass --tag claude-code and --session-id ${CLAUDE_SESSION_ID} so the capture carries the right tag and session attribution.

Capture one todo per keeper, not one blob. Each becomes an independently triageable backlog item. Don't re-capture the feed cards as new ideas; the digest already framed them. The local phase adds the X-only cards and tags everything.

Hand Off

Once the keepers are captured, the work belongs to improve-claude-code: it triages the claude-code backlog into batched PRs. Tell the user how many todos landed and that they can run improve-claude-code to process them.

Gotchas

  • No completion notification exists for routines. The doorway item is the signal. If its creation fails, say so loudly in the run output rather than ending silently.
  • Teleport, not resume. The handoff is claude --teleport <session>, which pulls the cloud session into a local terminal with local tools. --resume does not do this. If the browser or Things isn't reachable after teleport, you may not actually be in a local session. Confirm before assuming the fallback is broken.
  • Thin weeks are normal. Low-frequency blogs often post nothing in an 8-day window. A digest of 0-2 ideas is a valid outcome, not a failure. On the weekly routine, still drop the doorway item so the cadence is visible.
  • One tag, one backlog. The claude-code tag is the contract with improve-claude-code. Without it, the todo won't surface there.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.