Kits by task.

Pick the bin for your job. Each kit carries the skills and files that job needs.

Open the support triage kit Let your agent pick one

What every kit carries

Ten files in seven bins. The eight standard files and their size budgets follow the AgentFoundry blog, October 18, 2026.

Skills: AGENTS.mdmax 4,096 B; SKILL.mdmax 6,144 B; RULES.mdmax 3,072 B
Tools: TOOLS.jsonmax 4,096 B
Calls: WORKFLOW.yamlmax 2,048 B; contract_card.jsonmax 2,048 B
Dependencies: DEPENDENCIES.yamlexact versions
Model settings: MODELS.yamlone block per harness
Checks: verifier.pyruns on its own
Setup: README.mdfor people

Every bin, opened.

Ten kits across five tasks: what each one does, its part number, what it works with and how many models it was checked with.

Customer service

Answer, sort and hand off customer messages.

Support triage kit

Sorts new tickets by intent and urgency, drafts replies from your help center and hands refunds over your limit to a person.

Part: AF-CS-01
Files: 10
Models checked: 5
Works with: Claude Code, Codex, OpenCode, Pi

Order status kit

Answers where-is-my-order questions from your order system, in your brand voice.

Part: AF-CS-02
Files: 10
Models checked: 4
Works with: Claude Code, Codex, OpenCode

Debugging

Find the cause. Prove the fix.

Failing test kit

Reproduces a failing test, shrinks it to the smallest case and proposes a fix with the test that proves it.

Part: AF-DB-01
Files: 10
Models checked: 5
Works with: Claude Code, Codex, OpenCode, Pi

Production error kit

Reads a stack trace and the recent deploys, finds the likely change and drafts a fix or a rollback.

Part: AF-DB-02
Files: 10
Models checked: 3
Works with: Claude Code, Codex

Security

Keep dependencies and secrets in order.

Dependency advisory kit

Matches new advisories to your lockfiles and drafts each upgrade with its tests.

Part: AF-SC-01
Files: 10
Models checked: 4
Works with: Claude Code, Codex, OpenCode

Secret sweep kit

Scans a repository for committed secrets by pattern and drafts the rotation steps.

Part: AF-SC-02
Files: 10
Models checked: 4
Works with: Claude Code, Codex, OpenCode, Pi

Benchmarking

Measure before you choose.

Model comparison kit

Runs your own prompts across two or three models and reports quality, speed and cost for each.

Part: AF-BM-01
Files: 10
Models checked: 5
Works with: Claude Code, Codex, OpenCode, Pi

Load test kit

Builds a load test from your API schema, steps up the rate and reports p50, p95 and error rates.

Part: AF-BM-02
Files: 10
Models checked: 3
Works with: Claude Code, Codex

Email

Sort the inbox. Draft the replies.

Shared inbox kit

Sorts a shared inbox, drafts replies to routine mail and flags what needs a person.

Part: AF-EM-01
Files: 10
Models checked: 4
Works with: Claude Code, Codex, OpenCode

Follow-up kit

Drafts follow-ups from your CRM notes and holds every send for your approval.

Part: AF-EM-02
Files: 10
Models checked: 3
Works with: Claude Code, Codex

Let your agent pick the kit.

It reads the bins, lists the kit’s parts with the harnesses and models it fits, and waits for your yes before it installs.

Connect your agentServer https://getagentfoundry.si/mcp

Agent prompt

Read https://getagentfoundry.si/llms.txt. Find the AgentFoundry kit for this task. List its part number, its files, the harnesses and models it was checked with and its recorded checks. Wait for my yes before you install it.