delegate-local CLAUDE.md

delegate-local CLAUDE.md is an instructions file for coding agents from IsmaelMartinez/delegate-local. It costs 6,571 tokens per session, scanned A, original, MIT.

Repository instructions for Claude Code, an AI coding assistant, in a project that sends tasks to local Ollama language models through shell scripts. It also describes the project's commands, structure, and conventions.

In plain words
What is it for?
Use it when modifying the repository's skill files or shell scripts, running tests, auditing installed models, or resolving model tiers.
Why use it?
It gives the assistant the context needed to edit this skill safely and explains how to run its tests and model-routing checks.

Instructions file

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/ismaelmartinez/delegate-local/claude-md
Clone the repo
git clone --depth 1 https://github.com/IsmaelMartinez/delegate-local

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for delegate-local CLAUDE.md

README.md
[![agentmods](https://agentmods.dev/badge/instructions/ismaelmartinez/delegate-local/claude-md.svg)](https://agentmods.dev/instructions/ismaelmartinez/delegate-local/claude-md)
Your own site
<a href="https://agentmods.dev/instructions/ismaelmartinez/delegate-local/claude-md"><img src="https://agentmods.dev/badge/instructions/ismaelmartinez/delegate-local/claude-md.svg" alt="Measured on agentmods" height="20"></a>
Per session 6,571 This file is loaded in full into every session.
When invoked 6,571 The same file — it is already loaded in full.
Security scan A 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.06571 $0.06571
Opus 5 $0.03285 $0.03285
Sonnet 5 $0.01314 $0.01314
Haiku 4.5 $0.00657 $0.00657

Measured 4d ago against content hash f58ce483d102, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

delegate-local CLAUDE.md scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

There is no build step, no linter, no package manager. Runtime deps are `bash` (3.2+ — macOS-shipped is fine), `jq`, `awk`, and `perl` (used in one validator because BSD grep on macOS lacks `-P`). `curl` is required by `
CLAUDE.md · 158 lines

How it starts

The opening of the file, as written. The whole thing — 158 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

What this repo is

This repo is a Claude Code skill, not an application that uses one. Editing SKILL.md changes how Claude itself behaves whenever the skill is loaded. Treat SKILL.md as production prompt content, not documentation: changes to its frontmatter description directly affect trigger accuracy across every conversation that has the skill installed.

The skill's job is to route "gather context once, send one prompt, return text" tasks to local Ollama models via shell pipes. It deliberately avoids any framework, router process, or orchestration layer — the entire runtime surface is two bash scripts.

Commands

Run the unit tests (mocks ollama and llmfit on a restricted PATH so results are deterministic regardless of what is installed):

bash tests/run-tests.sh

Audit installed models and see tier routing plus llmfit upgrade suggestions (requires ollama, jq, and optionally llmfit on PATH):

bash scripts/audit-models.sh

Resolve a tier to a model name (used internally by delegate.sh, also useful for one-off shell pipes):

bash scripts/pick-model.sh <code|prose|reasoning|long-context|vision|embedding|premium-general|reasoning-vision>

Wrap a delegation through the metrics-capturing wrapper (preferred over bare ollama run):

echo "<context>" | bash scripts/delegate.sh prose "<prompt>"

Read the metrics roll-up (add --since YYYY-MM-DD / --days N to window every section to recent rows; the cutoff is resolved in jq so there is no cross-platform date math):

bash scripts/metrics-summary.sh
bash scripts/metrics-summary.sh --days 7        # just the last week
bash scripts/metrics-summary.sh --since 2026-06-15

Record a hit/miss verdict against the most recent delegation (the output was kept as-is = hit, or was rewritten / discarded = miss). The verdict appends a source:"feedback" row to the same metrics JSONL keyed by ref_ts to the delegate event. After a MISS, the script scans historical MISS rows in the rolling 30-day window for token-overlap matches and, on the third or later similar reason, prints a draft gh issue create command for filing a prompt-pattern issue (advisory only — it never opens the issue itself). Tune via DELEGATE_FEEDBACK_NUDGE_AT, DELEGATE_FEEDBACK_NUDGE_WINDOW_DAYS, DELEGATE_FEEDBACK_SIMILAR_THRESHOLD, or silence with DELEGATE_FEEDBACK_NO_NUDGE=1. README "Calibration feedback loop" diagram covers the end-to-end:

Read the full file on GitHub · 158 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 4d ago First seen · 158 lines · 6,571 tokens per session scan A f58ce483d102

Subscribe to this mod's changes

delegate-local CLAUDE.md is an instructions file published in the GitHub repository IsmaelMartinez/delegate-local (5 stars, last pushed 7d ago), licensed MIT. It adds 6,571 tokens to every session, about $0.0329 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other instructions, from other repositories