|
1 | | -# Strategic plan: turn skill creators into the growth engine |
| 1 | +# Ready for Agent Plugins v1 (agent-plugins.org) |
2 | 2 |
|
3 | | -## Goal |
4 | | -Move the platform from a "tool subscription" model to a **creator-owned marketplace economy**. The primary lever is retention & activation of skill creators, with the hardest drop-off being users who see value but do not pay. The long-term bet is that creators who earn money from their skills will supply higher-quality capabilities, which attracts more end users, which drives more Pro conversions, which pays creators more. |
| 3 | +Agent Plugins is the new vendor-neutral package format (TSC from Amazon, Cursor, Microsoft, OpenAI, Vercel) that wraps two component types we already ship: Agent Skills (`SKILL.md`) and MCP servers. Verified spec v1.0.0 requirements: a plugin is a directory with a required root `plugin.json` (closed schema: `$schema`, `name`, `version`, `description`, `author`, `homepage`, `repository`, `license`, `keywords`, `extensions`), optional `skills/<name>/SKILL.md`, and optional root `mcp.json` (`$schema` + `mcpServers`, per-server `stdio` / `streamable-http` / `sse` variants). |
5 | 4 |
|
6 | | -## Current state (verified) |
7 | | -- Creators can author, evaluate and publish skills via `/forge` and get a Trust Score. |
8 | | -- The marketplace ranks by popularity first, Trust Score second, with a niche-penalty for stack-specific skills. |
9 | | -- The CRM already segments creators by lifecycle stage and can nudge them toward publishing and Pro. |
10 | | -- Value-proof reports translate score improvements into dollars and engineer-hours for end users. |
11 | | -- Pro is priced at `$19/mo` or `$140/yr` and unlocks batch reviews, the Agent Store and SAK University. |
12 | | -- There is no visible revenue-share or creator-payout mechanism tying creator earnings to platform usage. |
| 5 | +Our current state: `content/skills/*.yaml` is the source of truth, `scripts/build-skills-sh-mirror.mjs` already emits spec-correct `skills/<slug>/SKILL.md`, and we serve a hosted MCP endpoint at `/api/public/mcp`. We are one manifest layer away from being a conformant plugin publisher. |
13 | 6 |
|
14 | | -## Strategic initiatives |
| 7 | +## 1. Repo becomes a conformant plugin |
15 | 8 |
|
16 | | -### 1. Creator monetization (revenue share) |
17 | | -Introduce a transparent creator economy so publishing skills becomes a revenue activity, not just a distribution activity. |
| 9 | +- Add root `plugin.json` (`name: superagentskill`, version from `src/lib/version.ts`, homepage/repository/license/keywords, `author`) so `npx`-style clients that read Agent Plugins can consume this repo directly. The existing `skills/` mirror already satisfies the skills component location. |
| 10 | +- Add root `mcp.json` declaring our hosted server as `type: "streamable-http"` with `url: https://superagentskill.com/api/public/mcp` (no secrets in `headers` — auth stays client-managed, as the spec requires). |
| 11 | +- Extend `scripts/build-skills-sh-mirror.mjs` (or a sibling `scripts/build-agent-plugin.mjs`) to generate both manifests from the same content source, plus a `--check` mode wired into `npm run check:skills-mirror` and the validate-content workflow so manifests never drift from the platform version. |
18 | 12 |
|
19 | | -- **Pro attribution**: when a user upgrades to Pro within 30 days of installing or reviewing a creator's skill, attribute a share of that subscription to the creator. |
20 | | -- **Pack purchases**: existing `pack_purchases` and `package_purchases` tables can be extended with a `creator_royalty_cents` column and a monthly payout accrual. |
21 | | -- **Payout ledger**: new `creator_earnings` table tracking per-skill accrued earnings, paid/unpaid status, and Stripe Connect / manual payout records. |
22 | | -- **Minimum thresholds**: e.g. `$25` minimum payout, 30-day holding period to reduce refunds/chargebacks. |
| 13 | +## 2. Per-skill and per-agent plugin packages (download + API) |
23 | 14 |
|
24 | | -### 2. Creator analytics dashboard |
25 | | -Build a `/creator/dashboard` (or `/creator`) route that shows each publisher: |
| 15 | +- New generator producing one plugin directory per catalog item: `plugin.json` + `skills/<slug>/SKILL.md` (+ `references/` when the skill has examples), reusing the current serializer. |
| 16 | +- New endpoints under `src/routes/api/public/`: |
| 17 | + - `plugins/<slug>/plugin.json` and `plugins/<slug>/mcp.json` — manifests served from live database content. |
| 18 | + - `plugins/<slug>.zip` — the full portable plugin package (skill files + manifests), so any conformant client can install a graded SAK skill with no account. |
| 19 | + - `plugins.json` — an index of available plugins for discovery. |
| 20 | +- Agent Store items (soul + skills + playbooks) map naturally to a multi-skill plugin: one `plugin.json`, several `skills/*`, `mcp.json` pointing at our server for graded updates. Pro gating stays where it is today. |
26 | 21 |
|
27 | | -- Installs, reviews, executions and Trust Score evolution per skill. |
28 | | -- Estimated revenue, attributed Pro conversions, and pack sales. |
29 | | -- Which skills drive the most value (outcome proof) for end users. |
30 | | -- Benchmarks against top creators in the same vertical. |
| 22 | +## 3. Conformance validator |
31 | 23 |
|
32 | | -This addresses activation by making the value creators generate visible in money and distribution, not just abstract scores. |
| 24 | +- `scripts/validate-agent-plugin.mjs`: validates generated manifests against the v1 rules we must not violate — closed top-level fields, plugin `name` charset (`a-z0-9-.`, no `--`/`..`, alphanumeric ends, 1-64 chars), canonical `$schema` identifiers, plugin-relative paths starting with `./` and contained in the plugin root, HTTPS-only non-loopback MCP URLs, no credentials in headers. |
| 25 | +- Tests in `tests/agent-plugins.test.mjs` covering a valid manifest, each fatal violation, and the packaged skill layout. |
33 | 26 |
|
34 | | -### 3. Outcome-based marketplace ranking |
35 | | -Evolve the marketplace from "most installed" to "most value delivered". |
| 27 | +## 4. Communicate it (this is the point of "preparing") |
36 | 28 |
|
37 | | -- Add an `outcome_score` derived from real execution outcomes, review deltas, and value-proof reports. |
38 | | -- Blend `outcome_score` with popularity and Trust Score so high-quality, high-impact skills surface even if they are niche. |
39 | | -- Show "verified outcome" badges on skill cards (e.g. "saved avg $X/month for teams"). |
40 | | - |
41 | | -This helps end users find skills that actually work, which increases installs and Pro conversion. |
42 | | - |
43 | | -### 4. Creator success program (SAK University for publishers) |
44 | | -Turn the existing residency/university infrastructure toward creator education. |
45 | | - |
46 | | -- Tracks: "Build your first skill", "Get to Trust Score A", "Monetize your expertise", "Build an agent team". |
47 | | -- Credentials: verified creator badges and "Top earner" status. |
48 | | -- Office hours / async feedback: the existing feedback-request system can be expanded into a mentor queue. |
49 | | - |
50 | | -### 5. Team/collaboration for creators |
51 | | -Allow creators to collaborate on skills, agents and playbooks. |
52 | | - |
53 | | -- Organizations / teams with shared revenue splits. |
54 | | -- Co-authoring on packages. |
55 | | -- Private team registries (enterprise bridge). |
56 | | - |
57 | | -This expands the addressable audience from individual creators to agencies and internal platform teams. |
58 | | - |
59 | | -## What we will NOT do in this phase |
60 | | -- Add new social features (follows, comments) — low activation impact. |
61 | | -- Lower Pro price — the drop-off is value perception, not price. |
62 | | -- Build a custom payment processor — use Stripe Connect for payouts. |
63 | | - |
64 | | -## Success metrics |
65 | | -- **Creator activation**: % of published creators who return within 14 days. |
66 | | -- **Creator monetization**: number of creators with >$0 accrued earnings; median monthly creator earnings. |
67 | | -- **Pro conversion attributed to skills**: % of Pro upgrades with a skill install/review in the previous 30 days. |
68 | | -- **Marketplace quality**: average Trust Score of top 50 skills; % of installs from skills with outcome badges. |
69 | | - |
70 | | -## Suggested first implementation slice |
71 | | -1. Schema: `creator_earnings` table and `creator_payouts` table with RLS/grants. |
72 | | -2. Server function: `attributeProConversion` to record earnings when a Pro subscription starts. |
73 | | -3. Route: `/creator/dashboard` with earnings, installs and Trust Score charts. |
74 | | -4. Marketplace: add outcome badge to `PackageCard` and an `outcome_score` column to ranking. |
75 | | -5. CRM trigger: `creator_first_earning` to notify creators when they have accrued their first dollar. |
| 29 | +- Landing: extend the existing `OpenSkills` section into "Works with the open agent standards" — Agent Plugins v1, Agent Skills `SKILL.md`, and MCP — with a copyable install line and the differentiator kept as-is (the standards give distribution; SAK adds the Trust Score, adversarial pass rate and before/after proof). |
| 30 | +- `/docs`: new "Agent Plugins (v1)" block with the layout we emit, the manifest example, the package download URL, and the compatibility matrix row added to `INSTALL_ROUTES` in `src/lib/skills/open-skills.ts`. |
| 31 | +- `/marketplace/<slug>`: extra install option "Download Agent Plugin (.zip)" next to the existing commands. |
| 32 | +- `/how-it-works`, `/welcome`, `/agents.md`, `/llms.txt`: one line each declaring Agent Plugins v1 conformance and the plugin endpoint, since agents read those first. |
| 33 | +- Update `head()` descriptions on the touched routes; ship as a minor version bump via `scripts/bump-version.mjs` with a CHANGELOG entry. |
76 | 34 |
|
77 | 35 | ## Technical notes |
78 | | -- Use `createServerFn` for attribution logic; keep payout reads under `requireSupabaseAuth`. |
79 | | -- Use Stripe Connect for payouts; do not store bank details in the database. |
80 | | -- Ensure all new public tables follow the GRANT/RLS/policy pattern. |
81 | | -- Keep all user-facing copy English-only per project rule. |
| 36 | + |
| 37 | +- No database schema changes, no new secrets, no new dependencies (zip built with a small pure-JS store-only writer or an existing dependency if one is already present). |
| 38 | +- Manifests are generated, never hand-edited, from `content/skills/*.yaml` and the live registry, so the repo mirror and the API stay identical. |
| 39 | +- All copy stays English-only; illustrative numbers keep their projection labels. |
0 commit comments