roadmap(phase 6): advanced contribution operating layer

[JSONbored]

Parent roadmap: #127

Goal

Build the post-v1 advanced contribution operating layer after phases 0-5 are stable: replayable decisions, repo-specific scenario planning, MCP-native workflows, custom agent distribution, policy compilation, and outcome learning.

This phase should make Gittensory feel much more useful without turning it into the wrong product. It is still a deterministic control plane for contribution work, not a reward predictor or autonomous PR bot.

Non-goals

• No public reward predictor.
• No public leaderboard.
• No wallet, hotkey, or coldkey dashboard.
• No autonomous PR bot.
• No public raw trust score, private reviewability, or private scoreability output.

Research basis

• Gittensor OSS contribution scoring: https://docs.gittensor.io/oss-contributions.html
• Gittensor repository hyperparameters: https://docs.gittensor.io/repository-hyperparameters
• MCP server primitives: https://modelcontextprotocol.io/specification/2025-06-18/server/index
• GitHub Checks API: https://docs.github.com/en/rest/checks
• GitHub custom agents: https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-custom-agents

Phase 6 child epics

• feat(agent): add replayable decision snapshots #270 replayable decision snapshots.
• feat(agent): add repo-specific scenario simulator #271 repo-specific scenario simulator.
• feat(mcp): ship native resources prompts and output schemas #273 MCP-native resources, prompts, and output schemas.

Related cross-phase dependencies

These are native children of earlier phase epics, but they support the advanced operating layer once those phases are stable:

• feat(distribution): add Gittensory custom agent profiles #276 Gittensory custom agent profiles. Parent phase: roadmap(phase 5): ecosystem distribution #238.
• feat(settings): add repo policy compiler for focus manifests #277 repo policy compiler for focus manifests. Parent phase: roadmap(phase 3): repo owner intake console #236.
• feat(analytics): add outcome learning loop for recommendation quality #278 outcome learning loop for recommendation quality. Parent phase: roadmap(phase 4): adoption analytics and launch system #237.

Done when

• The phase owns advanced post-v1 work without creating another top-level roadmap.
• Implementation work is split into focused child issues that fit one contributor PR each.
• Advanced planning remains advisory and authenticated where needed.
• Public GitHub surfaces stay sanitized and low-noise.
• Implementation PRs include targeted tests plus npm run test:ci.

Cross-cutting acceptance criteria

• Public output must not expose wallets, hotkeys, reward estimates, raw trust scores, public score predictions, private reviewability, private scoreability, or farming language.
• MCP/local tooling must keep source-upload behavior fail-closed.
• Scenario and outcome systems must not claim guaranteed payout outcomes.
• Human approval remains required for issue filing, PR creation, comments, labels, closure, merge, and public GitHub actions unless a separate maintainer-approved policy explicitly allows a narrow action.