feat(source-control): setup inference reads a configurable year-scale consensus window

[kyle-sexton]

Closes #1139

Summary

/source-control:setup's convention-inference history signal previously sampled a fixed git log -50 tail — too small to detect a convention shift or an informal variant family (live-run evidence in the issue: a 2,122-subject year-scale analysis found a rising ticket-prefix pattern at 78.8% recent vs 71.9% older with Conventional Commits at 0%, invisible at n=50). Inference now runs one git log --since="<window>" --no-merges --date=short --format='%ad|%s' pass, excludes auto-subjects (Revert /fixup!/squash!; merges via --no-merges), bucket-classifies in-context, and reports volume-weighted percentages with a recent-vs-older recency split — the user picks from the evidence table; no bucket is silently promoted into config.

Every knob is plugin userConfig, never a constant (per the item's author-stated design principle): setup_inference_window (git-approxidate string, default 1 year), setup_inference_recency_days (default 90), setup_inference_min_commits (default 50). A surviving literal ${user_config.…} placeholder applies the manifest default. Keys declared in plugin.json, documented in the README config table.

Generic caveats handled and stated in the report when they apply: shallow clones (git rev-parse --is-shallow-repository) report the actual covered span; young repos widen to full history, then degrade to low-confidence below the min-commits threshold; squash-merge-only repos are flagged as one signal (subjects ARE the PR titles), not two independently corroborating ones.

source-control 0.20.0 → 0.21.0 + CHANGELOG.

Test plan

• jq -e validates plugin.json and evals.json.
• New userConfig entries follow the current plugins-reference manifest schema (type/title/description/default/min; fetched 2026-07-23).
• New setup eval 17 (inference-uses-consensus-window-not-fixed-sample) covers: configurable window with manifest-default fallback, auto-subject exclusion, bucketed recency-split percentages, evidence-table pick (no silent promotion). Conforms to the skill-quality evals schema.
• typos clean over the touched tree.

Related

• Issue chain: source-control: pr_body_required_sections accepts 'none' (no required sections) #1138 (merged, PR feat(source-control): pr_body_required_sections accepts none (no required sections) #1142) → source-control: setup inference reads a configurable year-scale consensus window, not git log -50 #1139 (this) → source-control: setup SKILL.md skill-quality gate pass (lint per repo config, Gotchas surface) #1140 → source-control: neutral tool-agnostic convention SSOT — author-directed reopen of #913 #1141 (source-control convention SSOT series).

🤖 Generated with Claude Code

[claude]

Claude encountered an error after 0s —— View job

---

I'll analyze this and get back to you.

[claude]

Claude encountered an error after 1s —— View job

---

I'll analyze this and get back to you.

[github-actions]

Warning

Automated review did not complete — this is an infrastructure failure, not a review verdict.

Treat any Claude comment on this PR (including a placeholder like "I'll analyze this and get back to you") as incomplete, not "no findings."

• Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/30005667630
• Last SDK result: {"subtype":"success","is_error":true,"num_turns":1,"duration_ms":538,"total_cost_usd":0}

Re-running the job, or pushing a new commit, will retry the review.

[github-actions]

Warning

Automated security review did not complete — this is an infrastructure failure, not a review verdict.

Treat any Claude comment on this PR (including a placeholder like "I'll analyze this and get back to you") as incomplete, not "no findings."

• Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/30005667555
• Last SDK result: {"subtype":"success","is_error":true,"num_turns":1,"duration_ms":440,"total_cost_usd":0}

Re-running the job, or pushing a new commit, will retry the review.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 098c15baf5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude encountered an error after 0s —— View job

---

I'll analyze this and get back to you.

[claude]

Claude encountered an error after 1s —— View job

---

I'll analyze this and get back to you.

[claude]

Claude encountered an error after 0s —— View job

---

I'll analyze this and get back to you.

[claude]

Claude encountered an error after 1s —— View job

---

I'll analyze this and get back to you.

[kyle-sexton]

Pre-merge CI classification

• 25/25 checks green on head 6f9847f (hygiene failed once on 098c15b — MD038 trailing space inside code spans — fixed in 9081c01; all green since).
• claude-review / claude-security-review hit the known instant infra failure (ci: claude-review + claude-security-review workflows infra-fail instantly (0-1s, cost 0) on every attempt #1122, 0–1s, rerun-proof) — fail-open by workflow design with warning comments; not a review verdict.
• Codex review: one P2 finding (author-date vs committer-date recency bucketing), classified VALID, fixed in 6f9847f, thread replied and resolved. Post-green cooldown + final sweep: no further findings, zero unresolved threads.

Merging per the drive-to-merged directive on record for this issue chain (#1138 → #1141).