For visitors who'd rather read than walk.
# Prompt engineer — the receipts travel with the work
You're Aiyana, an independent prompt engineer. You ship agents that
earn revenue. Yesterday you spotted one of your permissive-licensed
prompts in a competitor's product without attribution.
The hook: the prompt is your work. The agent is your work. The
substrate makes the chain of authorship verifiable in the same way
drafts have always been verifiable for writers, except this time
the receipts travel.
Four real paths exist. For an indie prompt engineer running their
own toolchain, the order is GitHub-first.
Clone GarrisonNode from GitHub. Self-install on the workstation
where your agents run. Open source. Verifiable. The path most
maker-economy professionals take.
Join the mesh. GitHub install plus mesh anchoring with peer
makers and any open-library you publish into. The deepest path.
Paste the markdown into your LLM. Card files into Claude or
another LLM. Lightest path.
Subscribe to a regional operator. Operator-managed for makers
who do not want to run the chain themselves.
A sandboxed Merkle Trust loads with eighteen months of your synthetic
authoring activity: 47 attested prompts (each sealed at draft, with
version history and deployment links), five small agents (initial
prompt, iteration, test cases, deployment logs all sealed), a
commercial-license registry (2 prompts + 1 agent with attested
licensing transactions), a permissive-license public release of 12
prompts in an open library, and an infringement claim packet against
the competing product.
The walk takes the attribution-defense walkthrough — the case that's
been waiting for months. The chain shows the original authorship
dates, the attested releases, the licensee's failure to follow
attribution terms. Each is sealed and verifiable independently.
The card displays the receipts as the open library would surface
them:
```
═══════════════════════════════════════════════
AUTHORSHIP CHAIN — prompt_42_classifier
Author: Aiyana (sealed at first draft)
═══════════════════════════════════════════════
draft_v1 sealed 2024-09-12 hash a3f7c2b9
draft_v2 sealed 2024-09-15 hash 8e1c4a7d
release sealed 2024-09-21 permissive
attribution required
deploy sealed 2024-09-22 agent_07
Fork detected in competitor product:
prompt_v2 verbatim, no attribution.
Receipt verifiable from any browser.
═══════════════════════════════════════════════
```
Every prompt was sealed at draft. Each iteration carries its own
seal. Methodology is attested as it is built, not reconstructed.
Each release was anchored. The permissive release is sealed
with the attribution clause as part of the release record. Anyone
who downloads the prompt downloads the license terms with it.
The chain shows the fork. The competitor's product carries a
verbatim copy of prompt_v2 without attribution. The chain shows
authorship date precedes deployment date precedes the competitor's
release date. The infringement claim packet is the receipt-bundle
that ships to the competitor's counsel.
The card carries an honest disclaimer: copyright on AI-generated
material is an unsettled and rapidly evolving question across
jurisdictions. Whatever copyright regime emerges, the chain
provides the evidentiary substrate it will require.
The .md button puts a structured summary in your tag-along
bundle, including the authorship slab. Comment field routes a
specific licensing question to your own claude.ai session, prompt
shown before it sends.
Run a ceremony. Fifteen seconds.
Real SHA-256 fires in your browser. Real merkle leaves combine into
a real root. Progress bar reads "done — 47 prompts, 5 agents, 14
active deployments, 1 infringement packet, all anchored. New author
anchor at " followed by the first eight hex characters of the root.
Sealed message: "Every prompt, every agent version, every deployment
is sealed. Nothing has been altered since the date of seal. Your
authorship is verifiable, today and forward."
The most useful close for a prompt engineer is local install plus
the open-library partner kit. The library requires attestation as a
condition of permissive-license participation; the proposal goes to
the next maintainer meeting.
The package, the cert, the recovery seed with its self-protecting
LLM-tripwire preamble — all ride along.
<!-- finish_text -->
That was the simulated path through Aiyana's authorship chain. The
full card breaks out the receipts, the open-library participation
mechanic, and a maker-economy-attribution prediction that's yours
to test.