agentleFS
BI-BlitzerBI-Blitzer/tandem-tokens/.claude/skills/offload/SKILL.md

offload

Skill1 starsChanged 4 months agoMITView on GitHub

Turn a bulk/mechanical data task into a validated ollamaq job spec, submit it to the local Ollama harness, and report the job id + ETA. Use when the user wants to clean/extract/classify/summarize many items, generate-and-test candidates, or run a staged pipeline — offloading the work off Claude's token budget.

5 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions5 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 5

    1. imbue-ai/offload★ 124same name
    2. borkweb/skills★ 8same name
    3. inkatze/planwright★ 4same name
    4. tide-tools/tide★ 1same name
    5. boxwright/offload★ 0same 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.

  • In this repository5 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 3

    Distinctive

    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. Distinctive:

    Follow these steps. The goal is a validated job spec that runs on a local model, so Claude tokens are spent only on the later reduce pass.

    line 8
  2. Distinctive:

    Check the policy in CLAUDE.md: ≥ ~20 similar independent items AND mechanical, generate-and-test, or staged.

    line 15
  3. Very specific:

    Write the spec to jobs/<name>.job.json (create jobs/ if needed).

    line 45
  4. Very specific:

    Small batch (< ~2 min): run it synchronously and go straight to results: ollamaq run jobs/<name>.job.json --config config/models.yaml …

    line 47
  5. Distinctive:

    If the spec is invalid, the command exits 2 with the exact validation error — fix the spec and retry.

    line 54

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “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

/offload — delegate bulk work to local Ollama

Follow these steps. The goal is a validated job spec that runs on a local model, so Claude tokens are spent only on the later reduce pass. See ../../../PLAN.md §3 for the full spec and [CLAUDE.md] for the when-to-offload policy.

1. Confirm it's worth offloading

Check the policy in CLAUDE.md: ≥ ~20 similar independent items AND mechanical, generate-and-test, or staged. If it fails the bar, say so and just do it inline.

2. Pick the mode and task_tag

  • map — one transformation per item (clean / extract / classify / summarize / codegen). task_tag ∈ clean, extract, summarize, classify, code.
  • gen_test — generate N candidates, score with a verifier, keep the best. task_tag: generate. Requires a verifier command and usually a loop.
  • pipeline — ordered stages; a per-stage filter decides which items continue. Remember: each stage must carry forward any payload the next stage needs (a classifier should output {flag, ...the data...}, not just {flag}).

3. Describe the input

input.kind: csv/jsonl (chunk rows), text (chunk lines/bytes), glob (chunk files, one file per item), or inline (values, chunk rows). Pick a chunk.size that fits the model context (start ~50 rows for cleaning, fewer for wide rows).

4. Write the prompt + schema

prompt.template MUST contain {{chunk}}. Add a JSON schema whenever the output is structured — Ollama constrains decoding to it, and the harness validates + repairs against it. Put "output ONLY JSON, no prose, no fences" in prompt.system. For gen_test, an optional {{exemplars}} placeholder gets the current top survivors fed back.

5. Emit and validate the spec

Write the spec to jobs/<name>.job.json (create jobs/ if needed). Then:

  • Small batch (< ~2 min): run it synchronously and go straight to results: ollamaq run jobs/<name>.job.json --config config/models.yaml --job-id <id>
  • Large / long / unattended: enqueue it for the daemon: ollamaq submit jobs/<name>.job.json --config config/models.yaml --job-id <id> then ensure a daemon is running (ollamaq status shows the queue; start one with ollamaq daemon if needed).

If the spec is invalid, the command exits 2 with the exact validation error — fix the spec and retry. Do NOT hand-wave past a validation failure.

6. Report

Give the user the job id, the resolved model, and a rough ETA from config/bench.json (that model's tok/s and load time — if bench.json is missing, suggest ollamaq bench). Then tell them to run /harvest (or check ollamaq status <id>) once it's done. Remind them that the map phase runs for $0 in Claude tokens; only the harvest/reduce pass will cost tokens.

Example map spec

{
  "name": "clean-contacts",
  "mode": "map",
  "model": "auto",
  "task_tag": "clean",
  "input": { "kind": "csv", "path": "data/contacts.csv",
             "chunk": { "strategy": "rows", "size": 50 } },
  "prompt": {
    "system": "You are a data cleaning engine. Output ONLY JSON.",
    "template": "Normalize these rows... return {\"rows\": [...]}. Input:\n{{chunk}}",
    "schema": { "type": "object", "required": ["rows"] }
  },
  "params": { "temperature": 0.1 },
  "retry": { "max_attempts": 3, "on_schema_fail": "repair_then_retry" }
}

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.