Agent adapter: read/write task lists via NIP-98 (agent-side counterpart to the browser app)

[melvincarvalho]

What

The tasks app gives a human add / complete / edit / delete over a wf:Tracker list in <pod>/public/tracker/. We built and proved an agent-side counterpart: a small skill that lets an autonomous agent do the same operations on the same resource, authenticated by the agent's own Nostr key via NIP-98 — no OAuth, no JWT.

Because it writes the identical wf:Tracker shape to the same /public/tracker/ location, the two interoperate: a task the agent adds shows up in the tasks app, and a task you add/complete in the app is visible to the agent. One tracker, two clients — human UI + agent. Not a new silo.

How it works (brief)

• The agent reads the list (public GET), edits the JSON-LD, and PUTs the whole document back with an Authorization: Nostr <base64 kind-27235 event> signature over its Nostr key.
• Two primitives (get / put) plus the agent's own reasoning give full CRUD — no per-verb endpoints. Add = append a Vtodo; complete = set status to COMPLETED; delete = drop it from issue.
• The auth mechanism and a few server-side papercuts are written up in jspod#79.

Question — where should it live?

This repo is currently a pure browser app (no Node code). The agent skill is small: a Node helper (NIP-98 signing + get/put) plus a short instruction file. Placement options:

1. An agent/ (or skill/) folder here, co-located with the app it mirrors — "tasks, for agents."
2. A separate package — since the signing primitive is generic pod read/write (not tasks-specific); only the wf:Tracker-shape knowledge is. That keeps this repo browser-only.

My lean: keep the generic NIP-98 read/write as a shared primitive elsewhere, and put only the thin tasks-specific instruction here (e.g. an agent/ doc + a tiny wrapper). But placement is the maintainer's call before any PR.

Size / shape

Tiny — ~100 lines across a helper + an instruction file. Currently ESM .mjs; would ship as .js inside a type: module package.

Refs: jspod#79 (the NIP-98 agent-write pattern).

[melvincarvalho]

Placement resolved: the generic skill + how-to now lives in its own repo — jsclaw/podwire (a NIP-98 read/write helper plus a "wiring an agent to a pod" guide).

Kept it generic, per the split floated above: the signing primitive isn't tasks-specific, so it shouldn't live inside this app. A thin tasks-specific instruction can reference it. The wf:Tracker interop point stands — an agent driving podwire writes the exact shape this app reads, so the two stay one tracker, two clients.