agentleFS
somasundarvsomasundarv/footprint/SKILL.md

footprint

Skill0 starsChanged 5 months agoMITView on GitHub

Reduce the number and size of tool calls (Read/Grep/Glob/Bash/Agent) and cut redundant context re-derivation. Changes the orchestration path, not the response style: fewer round-trips, narrower reads, subagent isolation. Use when the user says "reduce tool calls", "minimize context usage", "lean mode", "cut token usage", "/footprint", or when a task needs broad exploration across many files.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. Amanlabh/footprint★ 18same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No verify-reads: After a successful Edit/Write, do not Read the file back to confirm the change landed — the tool call already failed …

    line 49
  2. Distinctive:

    Delegate and isolate for anything wide: If a sub-question would take more than ~2 files or ~200 lines to answer, route it to a subagent …

    line 64
  3. Distinctive:

    Don't re-summarize prior findings in your own reasoning each turn — refer back to the ledger.

    line 73
  4. Distinctive:

    Don't preload what you don't need: If deferred tools are available (ToolSearch-gated), only resolve the schema for a tool right before you …

    line 76
  5. Distinctive:

    No speculative scope creep: Only touch what scope-lock (step 1) named.

    line 80
  6. Distinctive:

    After completing a task under this skill, report actual counts, not estimates:

    line 87
  7. Distinctive:

    Never fabricate a token/percentage figure.

    line 97

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Footprint: orchestration-level token optimization

What this is not

This is not a writing-style skill. It does not shorten sentences, drop articles, or compress prose. A response produced under this skill can be as verbose as normal. What shrinks is the scaffolding around the response: number of tool calls, bytes read per call, and repeated context re-derivation across turns. It can be paired with any output-compression or verbosity style — the two axes are independent and additive, not overlapping.

Protocol (apply in order, every task)

  1. Scope-lock before touching a tool. State the target in one line: file(s), symbol, or error string. If the request is ambiguous enough that you'd need to explore >3 locations to even find the target, ask ONE clarifying question (AskUserQuestion) instead of running a broad exploration pass. A wrong guess costs more tool calls than one question.
  2. Narrow before wide. Order of operations: Glob/Grep with a specific pattern → Read only the matched line ranges (offset/limit) → never Read a full file over ~300 lines speculatively. Never find or ls -R a whole tree when a targeted grep -r <symbol> answers the question.
  3. No verify-reads. After a successful Edit/Write, do not Read the file back to confirm the change landed — the tool call already failed loudly if it didn't. Only re-read if you need to reason about surrounding code you haven't seen yet, or a test/build step requires it.
  4. Batch, don't serialize. Any tool calls with no data dependency between them go in the same turn. If step 2 needs results from 3 independent greps, issue all 3 together, not one-then-wait-then-next.
  5. One reconnaissance pass. Gather everything you can predict you'll need in the first exploration round. Avoid the pattern of grep → read → "oh I also need X" → grep again → read again. Think one step ahead: if you're reading a function to understand it, also capture its call sites in the same pass if you already suspect you'll need them.
  6. Delegate and isolate for anything wide. If a sub-question would take more than ~2 files or ~200 lines to answer, route it to a subagent (Explore, or a project-specific investigator agent if available) instead of doing it inline. Only the digest comes back into the main thread — the exploration bytes never enter your context. This is the single biggest lever available; use it earlier than feels natural.
  7. Persist state, don't restate it. For multi-step work, use TaskCreate/TaskUpdate (or the project's todo mechanism) to hold the plan. Don't re-summarize prior findings in your own reasoning each turn — refer back to the ledger.
  8. Don't preload what you don't need. If deferred tools are available (ToolSearch-gated), only resolve the schema for a tool right before you call it, not speculatively for tools you might use.
  9. No speculative scope creep. Only touch what scope-lock (step 1) named. "While I'm here" edits or reads inflate both the tool-call count and the diff the user has to review. If you notice an unrelated issue, name it in one line at the end — don't go fix it.

End-of-task report

After completing a task under this skill, report actual counts, not estimates:

Footprint: N tool calls (R reads, G greps/globs, E edits, A agent
delegations). [If a naive baseline is being compared: state the naive path
explicitly and why it would cost more — don't assert a percentage without
showing the counterfactual path.]

Never fabricate a token/percentage figure. If you want to show savings, show the actual call count for the approach taken, and — only if you genuinely would have taken a broader path without this skill — name that alternate path and its call count so the comparison is falsifiable.

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.