Part of #1058 · Builds on / supersedes the intent of #816
Why
The loop's entry point is its weakest link. Today miner-side discovery is reactive: gittensory_get_issue_quality and decision_pack rank only within a repo the miner already watches, and gittensory_watch_issues is a dumb subscription. There is no cross-repo "what should I build right now," and no proactive alert when a high-value issue opens in a miner's lane. This is the biggest net-new gap on the miner side.
Current pieces to build on
src/signals/issue-quality.ts — per-repo issue ranking (actionable / needs-proof / stale / duplicate-prone / solved)
src/services/decision-pack.ts — per-miner opportunity pack (per repo)
src/services/agent-orchestrator.ts — agent_plan_next_work (already spans subscribed repos)
gittensory_watch_issues + list_notifications — an alert channel already exists
Requirements
- Aggregate opportunity signals across all registered repos, not just subscribed ones.
- Ranking must be deterministic and metadata-only — reuse existing signals; no new ML, no exposure of raw scores/rewards.
- Proactive alerts must be opt-in and filterable (lane / label / freshness).
- Respect the miner/maintainer privacy boundary at every surface.
Deliverables
Acceptance criteria
- A miner can ask "what's the best issue to build right now" and get a deterministic, cross-repo ranked shortlist via MCP.
- New high-value issues in a miner's lane generate an opt-in notification.
- No raw scores / rewards / hotkeys / wallet data exposed; ranking is reproducible from public metadata.
Out of scope
- The miner's own AI harness still does the building — this only surfaces what to build.
- Autonomous claiming/assignment of issues.
#816 (gittensor:feature, contributor-claimable) covers folding upstream-inspired advisor/scan ideas into miner planning. This issue supersedes its intent at the planning level but is currently labeled maintainer-only for cluster consistency. Decide whether to (a) close #816 as superseded, or (b) keep #816 as the contributor-eligible slice of this work.
Labels: roadmap, maintainer-only · Assignee: @JSONbored
Part of #1058 · Builds on / supersedes the intent of #816
Why
The loop's entry point is its weakest link. Today miner-side discovery is reactive:
gittensory_get_issue_qualityanddecision_packrank only within a repo the miner already watches, andgittensory_watch_issuesis a dumb subscription. There is no cross-repo "what should I build right now," and no proactive alert when a high-value issue opens in a miner's lane. This is the biggest net-new gap on the miner side.Current pieces to build on
src/signals/issue-quality.ts— per-repo issue ranking (actionable / needs-proof / stale / duplicate-prone / solved)src/services/decision-pack.ts— per-miner opportunity pack (per repo)src/services/agent-orchestrator.ts—agent_plan_next_work(already spans subscribed repos)gittensory_watch_issues+list_notifications— an alert channel already existsRequirements
Deliverables
issue-qualityacross all registered repos, ranked by reward-potential × lane-fit × queue-pressure × freshness (recency-weighted). New join over existing signals, not a new model.gittensory_find_opportunities(or extendagent_plan_next_workto span all registered repos, not just subscribed) — returns the ranked cross-repo shortlist, metadata-only, privacy-respecting (no raw reward). Wire into both MCP surfaces (hosted/mcp+ stdio npm package).Acceptance criteria
Out of scope
Note on #816
#816 (
gittensor:feature, contributor-claimable) covers folding upstream-inspired advisor/scan ideas into miner planning. This issue supersedes its intent at the planning level but is currently labeledmaintainer-onlyfor cluster consistency. Decide whether to (a) close #816 as superseded, or (b) keep #816 as the contributor-eligible slice of this work.Labels:
roadmap,maintainer-only· Assignee: @JSONbored