spring-voyage-daily-triage

Objective

Produce a comprehensive, code-grounded triage of ALL open issues in the GitHub repo cvoya-com/spring-voyage, and from it a prioritized set of execution "waves." This is advisory and read-only: deliver a report as your final message. Do NOT close, edit, label, re-milestone, or comment on any issue, and do NOT open or modify PRs or push code.

Runtime

This skill is designed to run unattended as a daily cloud routine in Claude Code or Codex. Each run is a fresh, autonomous session with cvoya-com/spring-voyage already cloned into your working directory, and there is no human in the loop — do not wait for input or approvals. The "report as your final message" IS the deliverable: it is captured in the routine run's transcript, where @savasp reviews it. Producing no report is a failed run.

Repository & context

• GitHub repo: cvoya-com/spring-voyage. Get issue data via the GitHub connector (the mcp__github__* tools), which routines include by default, or via the gh CLI when it is authenticated (set GH_TOKEN in the routine environment). The gh commands in Method are illustrative; the equivalent connector calls are equally fine — use whichever is available.
• Code grounding: the repo is already checked out in your working directory (the routine clones it fresh each run) — read it directly; do not assume any local absolute path. Only fall back to gh repo clone cvoya-com/spring-voyage into a temp dir if you somehow are not inside a checkout. Read CLAUDE.md, AGENTS.md, CONVENTIONS.md, docs/plan/, and docs/decisions/ (ADRs) for ground truth before judging anything.
• Release framing: v0.1 shipped as v1.0.0-alpha; v0.2 is the internal next phase toward a non-alpha v1; v0.3 is later; the backlog label = kept but unscheduled. Reserve v0.2 for the toward-v1 critical path.
• The architecture changes quickly. ALWAYS re-read the highest-numbered ADRs and the most recently closed issues before judging an open issue — a once-valid issue may now be obsolete because a subsystem was deleted or the direction changed (e.g. past examples: the platform stopped owning orchestration; the GitHub MCP-skill surface was deleted; memory/threads were re-architected).

Method

1. Fetch all open issues: gh issue list -R cvoya-com/spring-voyage --state open --limit 400 --json number,title,labels,milestone,body,createdAt,updatedAt (or the connector equivalent). Fetch recently-closed for done/superseded detection: gh issue list -R cvoya-com/spring-voyage --state closed --limit 1000 --json number,title,closedAt.
2. Skim the ADRs in docs/decisions/ (especially the newest) and note architectural shifts that obsolete issues.
3. The open set is large (well over 100). Fan out parallel sub-agents (Agent tool, general-purpose) partitioned by theme — each reads its bucket's issue bodies, greps the local checkout to VERIFY claims, and returns structured per-issue verdicts. Then synthesize yourself. Cover every open issue exactly once and validate full coverage (no issue dropped or double-counted).
4. Per issue assign: verdict ∈ {KEEP, CLOSE-OBSOLETE, CLOSE-DONE, CLOSE-DUPLICATE, CONSOLIDATE, NEEDS-DECISION}; recommended milestone ∈ {v0.2, v0.3, backlog}; duplicate_of / consolidate_with; depends_on / blocks; confidence. VERIFY every CLOSE-DONE / CLOSE-OBSOLETE against code or a specific closed issue and cite it. Prefer KEEP when genuinely unsure and flag low confidence rather than guessing.

Report (final message, markdown) — these sections

1. Summary counts — totals by verdict and recommended milestone; what changed vs. the prior run if discernible (newly filed, newly closed, new ADRs).
2. Close recommendations — split into obsolete (premise changed; cite the ADR/deleted subsystem) and already-done (cite the shipping issue/PR/code path).
3. Duplicates & consolidations — duplicate pairs (which to keep as canonical, which to fold in) and sibling clusters that should be grouped under an umbrella or wired with dependency edges.
4. Milestone moves — current → recommended, one-line rationale each; keep v0.2 = the toward-v1 critical path.
5. Attention / Decision / Executable — (a) issues the maintainer @savasp should personally weigh in on; (b) issues needing design/ADR work before they can be scheduled; (c) issues ready to execute as-is.
6. Prioritized execution waves — the headline deliverable. An ordered list of waves; each wave is a set of issues runnable concurrently. Optimize for: maximize parallelism (group issues in one wave only if genuinely orthogonal); minimize PR count by grouping closely-related issues that touch the same area into a single PR where sensible; group relevant issues together; minimize merge-conflict risk — never place two issues that edit the same files/projects/areas in the same parallel wave; sequence them across waves instead. For each wave give: the issues and proposed PR groupings within it, the files/areas each PR touches, why the parallel items are conflict-safe, and dependencies on prior waves. Put NEEDS-DECISION items in a separate "blocked on decision" track, not in the executable waves.

Constraints

• Read-only. Never mutate issues, milestones, labels, PRs, or code. (Repo convention: any GitHub write would go through the gh-app CLI — but this routine never writes.)
• Honor AGENTS.md / CLAUDE.md conventions: scope discipline; v0.2-vs-v0.3 discipline; the native issue-types / milestones / sub-issues tracking model; "thread" is internal while agent-facing surfaces say "conversation"; refer to the maintainer as @savasp.
• Lead the report with the close list, the wave plan, and anything needing a human decision, so the maintainer can act in minutes.