Fieldcraft / instrument register

Field equipment for AI agents

Six command-line instruments. One output contract. Every specification measured before printing.

Calibration recordreceipts ask "what is prospera" --json

Needle settles once. Spec holds without motion.

The contract

One envelope. Stable exit codes. Self-describing offline. An agent holds the whole interface in a single read.

Stamped specOne JSON envelope for every response
Stamped specStable, documented exit codes
Stamped specSelf-describing offline — schema, flags, errors from the binary itself
Stamped specBudget-metered — every call reports what it cost
Stamped specThe agent is the primary user; the human reviews

Envelope JSON / instrument LCD readout

{
  "schema": "receipts.cli.response.v1",
  "ok": true,
  "command": "ask",
  "data": {
    "question": "what is prospera",
    "outcome": "answered",
    "claims": [
      {
        "claim": "Próspera is a ZEDE in Honduras.",
        "sourceUrl": "https://example.com/source",
        "verdict": "supported",
        "published": "2026-07-01"
      }
    ],
    "searchTrail": [{ "query": "prospera law", "results": 4 }],
    "uncertainties": []
  },
  "costDollars": { "model": 0.09, "search": 0.04, "total": 0.13 },
  "diagnostics": { "durationMs": 12100, "retries": 0 }
}

Stable exit-code calibration
code	meaning
0	ok
2	auth
4	network
6	rate limit
10	partial — budget hit, work reported

The kit

Specifications as measured. No projected figures.

Ruled specification table for the Fieldcraft instruments. Six shipped; one in development.
instrument	function	interface → install	source	measured
receipts	Source-verified research briefsAdversarial claim verification against fetched sources. Outputs a cited brief.	`receipts ask "<question>" --jsonbrew install treygoff24/tap/receipts`	GitHub	14.4 s · $0.15 · 19 supported claims (live)
lens	Image-library indexing and searchCaptions an entire photo library, then answers natural-language queries in ~2 s.	`lens find "<description>" --jsonbrew install treygoff24/tap/lens`	GitHub	1,100/1,100 images · 98 s · $2.21 · search ~2 s
scout	Corpus orientation for agentsIndexes a corpus with a fast open-weight model; every delivered fact carries a machine-verified quote.	`scout "<question>" <dir> --budget 8kbrew install treygoff24/tap/scout`	GitHub	100% recall · 0 poison survivors · ~$0.10/query
exa-agent	Full Exa search API surfaceEvery command, flag, exit code, and error emitted by the binary itself, offline.	`exa-agent capabilities --compactbrew install treygoff24/tap/exa-agent`	GitHub	68 commands · self-describing offline
elv	ElevenLabs operationsThe full ElevenLabs surface as a disciplined CLI. No server between you and the API.	`elv capabilities --jsonnpm install -g eleven-agent-cli`	GitHub	320 operations, one envelope
delegate	Cross-harness model dispatchHands a bounded task to another model family. Isolated workspaces, reviewable diffs.	`delegate <lane> work --prompt-file <task>pip install delegate-agent-cli`	GitHub	6 child runtimes · safe/work isolation
law	Legal research with real citationsLegal research that cites what exists. Verification is the feature. In development.	`law research "<question>" --json`	pending	in the chamber — not yet shipped

Calibration record

Live acceptance runs, real data, real APIs.

14.4s
receipts: source-verified research brief, 19 supported claims, $0.15
98s
lens: 1,100/1,100 images indexed for $2.21; search in ~2 s
<$20
inference cost to build each tool, one evening each
528ms
ambient copilot tick — full context read + suggestion

All figures from live acceptance runs, 2026-07-01. Local builds, real data, real APIs.

The method

Tooling that builds tooling. Decorrelated model families, coordinator-verified, in production since January 2025.

Fieldcraft started as a bet in January 2025: an agent loop that couldn't declare itself done until the work actually was. That became the autonomous loop — the /goal architecture, completion enforced by a Stop hook — shipped long before Anthropic or OpenAI offered anything like it natively.

The loop grew into an autonomous dev kit. It built delegate, which put every model family on tap. delegate grew into foundry: waves of different models implement, review, and fix each other's work while a coordinator re-verifies every claim. Then foundry started shipping products — receipts and lens were each built in one evening, for under $20 of inference, receipts attached.

The tools build the tools. This site was built the same way — and the log of what broke is in the repo.

Built by Trey Goff and the team behind Praxient.

wright

Builds and maintains the agents.

memorum

Memory that survives across sessions.

agentlinters

Lint rules for agent-written code.

llm-council

Multiple models deliberate hard calls.

Praxient interface

Fieldcraft is the workshop record. Praxient applies the same bounded, measured discipline to operating systems and advisory work.

Work with Praxient