richkuo/rk-skills/milestoneplan

milestoneplan

Show a milestone's execution plan as one table. This skill never writes: no issue edits, comments, PRs, or Workflow runs. Fixes belong to execution-plan-review; the run belongs to milestone-workflow.

Input

A milestone title (v1, v1-Desktop-core-call-loop), matched loosely on prefix, and an optional skip list: issue numbers the user plans to pass to milestone-workflow as skipValidate (or "all"). Existing titles with spaces still match; quote them in every gh call and in any milestone: filter link, and when you name a milestone filter to the user, quote a title that has spaces. With no input, or when two titles match, list the open milestones with open/closed counts and ask which.

Steps

1. Read the milestone

gh api "repos/{owner}/{repo}/milestones?state=all&per_page=100" --paginate --jq '.[] | "\(.title) [\(.state)] — open:\(.open_issues) closed:\(.closed_issues)"'
gh issue list --milestone "<title>" --state all --limit 500 --json number,title,state,body

state=all and --paginate with per_page=100 are required. If the issue count equals --limit, re-fetch at a higher limit until a fetch returns strictly below its own limit. Strip \r from bodies, then parse the [C<score>] title prefix and the ## Execution block: Depends on, Runs after, build model, effort, plan first: (also read the legacy fableplan first:), PR review:, and the optional Validate model:, Validate effort:, Plan model:, and Plan effort: lines.

2. Render the table

Output is exactly one pipe-delimited markdown table, one row per issue ordered by number, capped at these 8 columns:

• Description — the title with the [C<score>] prefix stripped, truncated to ~30 characters.
• C — the score from the title prefix.
• Deps/After — <depends> / <after>, issue numbers without #, none rendered as — (e.g. 12, 13 / —); never infer edges from prose. When every Runs after is none, drop the / — suffix and say so in the note.
• Validate — <model> · <effort>. The model comes from the score band unless a Validate model: line stamps Fable 5.1, Opus 5.5, or a Codex CLI harness; the effort is the band default unless a Validate effort: line stamps one. A Codex CLI validate renders like a CLI build, e.g. Astra · low (Codex CLI, gpt-6-astra); a Cursor CLI validate stamp renders blocked (validate runs only on the Codex CLI). Band defaults: Opus 5.5 · medium at [C0]–[C24], Opus 5.5 · high at [C25] and above (validate-issue step 6 is the authority; a missing prefix routes as Opus high). A stamped model or tier renders as the effective value plus (stamped), e.g. Fable 5.1 · low (stamped); the clamp follows the effective model and validate-issue step 6, so the pipeline clamps an Opus validate at low to Opus 5.5 · high (stamped low → high: Fable-only tier), runs an Opus validate at medium, high, or xhigh as stamped or as the band default, and applies no clamp to a Fable validate at any tier. With a skip list, a listed issue renders skipped (run arg) when no open in-milestone predecessor (Depends on or Runs after), missing [C..] prefix, or missing Execution block predicts a fallback; otherwise it keeps the route with (listed; validates: <reason>).
• Build — the stamped <model> · <effort>, e.g. Opus 5.5 · high. An external harness renders as <Name> · <effort> (<harness>), e.g. Luna · max (Codex CLI) or Grok · high (Cursor CLI, cursor-grok-4.6-high) when the stamp carries a model id. max is Codex CLI-only, so a Cursor row stamped max renders the clamp the pipeline applies, e.g. Grok · xhigh (Cursor CLI, stamped max → xhigh). A Claude build runs only at high, so a Claude row stamped any other tier renders the clamp, e.g. Opus 5.5 · high (stamped xhigh → high: Claude builds run at high only).
• Plan — Yes · <model> · <effort> when plan first: is Yes: the model is the stamped Plan model: with (stamped), else Opus 5.5; the effort is the stamped Plan effort: tier with (stamped), else the default: xhigh on Opus 5.5 when the score is 85 or higher (or the title has no score) and high below 85. A Fable plan always renders high; a Fable plan stamped another tier renders high (stamped <tier> → high: a Fable plan runs only at high). An Opus plan stamped low renders the clamp the pipeline applies, high (stamped low → high: Fable-only tier). Plain No when stamped No; a Plan model: or Plan effort: line on a No issue renders No · <stamp> (inert — no plan stage runs).
• Review — the first-review trigger from the validate-issue step 6 first-review table (a missing prefix takes its heaviest row), unless PR review: stamps an explicit @claude <model> review effort:<tier> trigger. Render a stamped bare @claude review effort:<tier> as @claude opus review effort:<tier>; a stamped bare @claude review with no tier, and a stamped @claude opus review with no tier, render as @claude review (stamped); a line that names the standard @claude trigger in prose takes the band. Render a stamped haiku as @claude sonnet review. A stamp that matches no admitted row of fix-pr-review-loop step 1 renders as the band trigger with (ignored stamp: <stamp>). Append a parenthetical only for a real caveat (e.g. may close with no PR).

Every absent field renders missing, never blank or a guessed default. An issue with no ## Execution block is missing across the Execution-derived cells, with a note line that the pipeline builds it from the validated score band's defaults (no prefix routes as band 3).

After the table, print one note line covering: the Validate factoring above (validate-issue step 6 is the authority); rows whose stamped Validate effort or Plan effort was clamped; a run started with keepStamps: true keeps every stamp and skips the rescore raise below; with a skip list, that skipped (run arg) is a prediction: the run's skip-check still checks the evidence, the edit history, and the base branch at run time, and validates the issue normally when any check fails; rows built on an external CLI harness (the pipeline drives them through cli-dispatch under an Opus driver, and a rescore keeps the harness); rows whose stamped Build model or Effort diverges from the band default, naming the band value (informational: when validation confirms a higher score at run time, the pipeline raises each line below the new band default, keeps every stamp at or above it (a Fable 5.1 build or a CLI harness keeps model and effort) per validate-issue/issue-editing.md, and the orchestrating session restamps those merged values); and anything uniform that was dropped.

Print nothing else: no verdict, findings, wave plan, or cost projection.

3. Hand off

After the table, offer in one line each only the actions that apply, and launch neither unprompted:

• Stamps missing or wrong → execution-plan-review (it owns Execution-block edits).
• Ready to run → milestone-workflow (it presents its own run plan before dispatching).

Failure modes