Repository navigation
[Epic] SkillX Marketplace & Agent-Native Platform Roadmap #31
Description
Goal
Evolve SkillX from a discovery directory into an agent-native skills/hooks marketplace with immutable packages, compatibility-aware discovery/install, personal kits, multi-device sync, credits-based commerce, and creator publishing.
Architectural decisions
- Keep AgentKit and SkillX independent.
- Do not port AgentKit's Go installer to TypeScript.
- AgentKit publishes through normal SkillX CLI/CI publisher workflows.
- Compatibility is machine-readable and evidence-backed.
- Agent-native spans CLI, MCP, WebMCP, Markdown routes,
llms.txt,llms-full.txt, and SSR. - Paid launch follows protected immutable delivery and entitlements.
Sequencing principle
First make packages trustworthy and reproducible; then installable; then executable hooks safe enough to distribute; only then attach money and creator payouts.
Child issues
Implementation order is encoded by the Depends on references in each phase body.
- Phase 0 — Foundations and protected data boundaries — [Phase 0] Foundations and protected data boundaries #32
- Phase 1 — Agent-native discovery and compatibility — [Phase 1] Agent-native discovery and compatibility #33
- Phase 2 — Immutable registry and publishing CLI/CI — [Phase 2] Immutable registry and publishing CLI/CI #34
- Phase 3 — Personal kits, installer, lockfile and multi-device sync — [Phase 3] Personal kits, installer, lockfile and multi-device sync #35
- Phase 4 — Hook packs and AgentKit reference publishing — [Phase 4] Hook packs and AgentKit reference publishing #36
- Phase 5 — Credits, orders, entitlements and paid launch — [Phase 5] Credits, orders, entitlements and paid launch #37
- Phase 6 — Creator economy, moderation and settlement — [Phase 6] Creator economy, moderation and settlement #38
Early non-goals
- Port AgentKit installer to TypeScript
- Make AgentKit depend on SkillX
- Arbitrary third-party installer plugins with full privileges
- Universal hook abstraction hiding runtime differences
- Automatic AgentKit/SkillX license-account merger
- Nested multi-publisher royalty graphs
- Autonomous spending without transaction-specific mandates
- Enterprise fleet management
- Premature microservice decomposition
Definition of success
- Creators publish immutable skills/hooks through stable CLI/CI.
- Compatibility is explicit and evidence-backed enough for agents to decide.
- Users compose kits and restore them reproducibly across devices.
- SkillX is usable through web, Markdown, CLI, MCP, and browser-agent tooling.
- Protected packages stay protected on every surface.
- AgentKit participates as a premium publisher without installer dependency.
- Paid transactions create auditable entitlements through a robust credit ledger.
- Third-party creators can sell under an approved settlement model.
Activity
zuey-cloud-harness commentedon Oct 11, 2026
Cross-repo AgentKit v3 plan
Main AgentKit epic: https://github.com/bestagentkits/agentkit/issues/2274; contract/ADR https://github.com/bestagentkits/agentkit/issues/2275. Workstreams:
Keep this existing roadmap as SkillX phase owner, not a competing registry. AgentKit owns Go installer, SkillX owns cloud registry/collections. Local G0 works offline; G1–G4 independently opt-in and cannot block v3 stable without explicit gate review.
mrgoonie commentedon Oct 13, 2026
Owner rollout steps done (2026-10-13):
main.skillx-dbbackup: a time-travel bookmark was taken before the migration (recorded in the owner's local env file). The rollback SQL is in the header of0011_release_tree_digest.sql.0011_release_tree_digest.sqlto the remote database. Verified that the newpackage_releasescolumns exist.skillx-webfrommainat beafa56, as version7889a09c-1a95-4c5f-ab8d-4bc60785070c.skillx.shand workers.dev both return 200.SKILLX_REGISTRY_SIGNING_KEYandSKILLX_REGISTRY_SIGNING_KEY_ID. The private key exists only in Cloudflare and in the owner's gitignored env file.Published registry signing key:
skillx-reg-993cd14a011f7b76D76gfU7S7-yl9ubULT4FQP8ocJgXES46rzY_gp9hG-ANot done yet: no runtime code reads the key so far. Signing the release manifest and status documents lands in SX-3, and the supervised publish/consume run follows that.
mrgoonie commentedon Oct 13, 2026
SX-3 (release pack/publish pipeline, flag off) merged in #59 as ec8e143. Next: SX-4 pinned read API (decide #60 first).
zuey-cloud-harness commentedon Oct 13, 2026
AgentKit v3 × SkillX cross-repo progress snapshot, 2026-10-14 00:25 ICT
Estimated combined implementation: ~30% by weighted bounded merged slices; 0/5 gates GO. This measures the AgentKit cross-repo epic #2274, not all phases or commerce work in this SkillX roadmap #31. Weights: G0 30%, G1 30%, G2 20%, G3 10%, G4 10%. G0=3/7(43%), G1=8/14(57%), G2/G3/G4=0.
SkillX G1 source milestones merged: SX-1 #57 (TS canonical tree digest), SX-2 #58 (signed release/status + migration 0011), SX-3 #59 (deterministic pack/R2/scoped publish + migration 0012), SX-4 #64 (pinned full-tree read), SX-5 #66 (audited state transitions + migration 0013/94-file E2E). 5/5 code slices merged, but SX #54 stays OPEN:
SKILLX_RELEASES_API=0, remote D1 migrations 0012/0013, activation and supervised real signed publish/consume remain undone. Follow-ups #65 (web CI), #68 (rate-limiter and real-D1), #67 identity selector, #63 Windows CLI test, #61 scanners.AgentKit consumer progress: AK-1 #2334 and AK-2 #2338 merged; AK-3..5 pending, local router RT-1 #2312 merged and RT-2 #2314 OPEN. Native G0 preflight #2307 is OPEN draft; adapter-projection #2308 blocks default-on. AK G2 #2278, G3 #2279 and G4 #2281 remain unimplemented. Full acceptance AK #2282 still NO-GO.
Immediate owner-run sequence: fix SX #65/#68 and migration 0012/0013+backup, deploy release API only after real D1 test, then scoped real signed publish/consume; concurrently finish AK #2307/#2308 and AK-3 read-only CLI integration. Keep private/sandbox execution disabled until separate gates pass. Epic detail & formula: https://github.com/bestagentkits/agentkit/issues/2274.
zuey-cloud-harness commentedon Oct 14, 2026
Cross-repo progress delta since the 00:25 ICT snapshot (2026-10-14 10:59 ICT)
AgentKit v3 x SkillX weighted merged-slice progress 30% -> 32%; G1 8/14 (57%) -> 9/14 (64%), driven by merged AK-3 #2339. G0 3/7 (43%) unchanged; G2/G3/G4 = 0%; all release gates NO-GO.
SkillX #69 and #70 merged. Maintainer confirms migration 0012/0013 and production
SKILLX_RELEASES_API=1, with anonymous endpoint smokes; #68 closed. New blocker #71: there is no publisher provisioning pathway/rows, so 0 releases exist and supervised signed publish/consume cannot run. Keep #54 open.Prioritize #71 + real signed release smoke, SX #65 CI; concurrently AK #2307 collision preflight and #2308 native projection coverage, then AK-4/AK-5 and RT-2..RT-4. Do not conflate enabled API or PR merges with end-to-end package/sandbox readiness. Full report is in https://github.com/bestagentkits/agentkit/issues/2274.
zuey-cloud-harness commentedon Oct 14, 2026
Cross-repo snapshot 2026-10-14 21:40 ICT
AgentKit v3 × SkillX estimated 43% (+11 points vs 10:59) by unchanged weighted merged-slice model: G0 4/7 (57%), G1 12/14 (86%), G2/G3/G4 0%. 0/5 gate GO, quality/production acceptance deliberately distinct. Source of truth: https://github.com/bestagentkits/agentkit/issues/2274.