claude-dev-env 4.3.0 → 7.0.0

package/.agents/skills/hitl/SKILL.md

@@ -1,45 +0,0 @@
----
-name: hitl
-description: >-
-  Maintain human-in-the-loop scope reporting; bound worker orchestration. Triggers: /hitl, HITL, human in the loop, out-of-scope work, scope update, worker status, Sol worker, Claude worker, Grok worker.
-disable-model-invocation: true
-user-invocable: true
-argument-hint: "[optional task context]"
----
-
-# HITL
-
-Initialize these human-in-the-loop rules for the active agent when `/hitl` is invoked.
-
-## Principle
-
-Keep the human informed when work outside the stated scope appears. Name the item, say why it is outside scope, and state the path taken.
-
-## Gotchas
-
-- Do not silently expand scope.
-- A worker's availability does not change the user's scope.
-- Sol, Claude, and Grok workers follow these same rules.
-- Preserve the requested work when an out-of-scope item appears.
-
-## When this applies
-
-Apply this skill only after explicit `/hitl` invocation. Treat any argument as context. Do not use it to add work to the user's request.
-
-## Process
-
-1. Read the active user scope before each material action.
-2. Keep work inside that scope.
-3. If an out-of-scope item appears, pause the affected path, report the item and the path taken, and ask for direction when needed.
-4. Use a low-effort Sol, Claude, or Grok worker only when the work stays in scope or the human has received the needed status. Give the worker this same scope rule.
-5. Report the worker choice and outcome, then continue the stated task.
-
-## File index
-
-| Path | Purpose |
-|---|---|
-| `SKILL.md` | Explicit `/hitl` operating prompt. |
-
-## Folder map
-
-`hitl/` contains only `SKILL.md`.

package/.agents/skills/pr-cleanup/SKILL.md

@@ -1,140 +0,0 @@
----
-name: pr-cleanup
-disable-model-invocation: true
-description: >-
-  Refine pull requests through parallel placement and capability-name audits,
-  focused delivery sizing, and a final simplify and code-review loop. Triggers:
-  /pr-cleanup, run PR cleanup, full PR cleanup, extraction audit, capability
-  naming audit, simplify and review a PR, and split a cleaned PR.
----
-
-# PR cleanup
-
-## Contents
-
-- [Principle](#principle)
-- [When this applies](#when-this-applies)
-- [Composition](#composition)
-- [Task seeding](#task-seeding)
-- [Process](#process)
-- [Promotion gates](#promotion-gates)
-- [Finish report](#finish-report)
-- [File index](#file-index)
-
-## Principle
-
-One coding agent owns the cleanup outcome. `pr-refinement` coordinates parallel
-preflight audits and produces findings, tested proposals, a combined change map,
-and a delivery decision for the cleanup owner. Parent-to-child promotion uses
-exact commit ancestry and fresh child-head checks.
-
-## When this applies
-
-Use this skill for a pull request that needs placement review, capability
-naming, cleanup convergence, and a focused delivery boundary.
-
-Required input: a pull request URL, number, or branch. If the target is missing,
-respond exactly: `Give a GitHub PR number, URL, or branch for pr-cleanup.`
-
-Use the repository that owns the target pull request. Keep every pull request in
-draft state until its applicable Ready gate is complete. Keep merge authority
-with the user.
-
-## Composition
-
-| Skill | Role | Evidence |
-|---|---|---|
-| `pr-refinement` | Run the parallel audits, combine findings, and coordinate implementation shape | Change map and delivery decision |
-| `pr-shared-extraction` | Find reusable behavior that belongs in `shared_utils` | Placement findings and tested proposal |
-| `pr-name-by-capability` | Find driver or motive words on reusable capability surfaces | Naming findings and rename directions |
-| `pr-small-cl` | Choose one coherent pull request or an ordered replacement stack | Focused boundary and dependencies |
-| `source-command-sr-loop` | Run `e-simplify`, then `e-code-review low --fix` until clean | Review passes, fixes, and validation |
-| `pr-summarizer` (repo-local) | When the repository under cleanup ships `.claude/skills/pr-summarizer/`, run it after Ready and post the secret-gist preview comment | Preview URL and comment confirmation |
-
-## Task seeding
-
-At skill start, register every item in `reference/task-seeds.md` as a session
-task through `TaskCreate`, `TodoWrite`, or the host task equivalent. Work from
-that task list. Mark each task complete with `PASS`, `FAIL` plus file and line
-evidence, or `N/A` plus the reason.
-
-## Process
-
-### 1. Resolve the target
-
-Resolve the pull request, repository, parent head SHA, and intended child
-boundary. Record the immutable parent preflight SHA before creating worktrees.
-
-### 2. Run `pr-refinement`
-
-Run [pr-refinement](../pr-refinement/SKILL.md). Record its combined change map,
-audit findings, locations, priorities, destinations or rename directions,
-validation evidence, and worker worktrees.
-
-### 3. Choose the delivery shape
-
-Use [pr-small-cl](../pr-small-cl/SKILL.md) after the audit findings are
-combined. Record the first pull request boundary, dependencies, tests, and
-follow-up work.
-
-### 4. Implement the selected shape
-
-Apply every actionable finding in dependency order. Keep preflight parent scope
-read-only until the cleanup owner selects and reapplies tested proposals.
-Validate each changed surface with its production-path tests. Commit each
-validated concern and keep the resulting pull request in draft state.
-
-### 5. Run `source-command-sr-loop`
-
-Run [source-command-sr-loop](../source-command-sr-loop/SKILL.md). Record the
-review passes, fixes, skips, tests, and commit SHAs.
-
-### 6. Promote and report
-
-After the applicable gate passes, when the repository under cleanup includes
-`.claude/skills/pr-summarizer/SKILL.md`, load and run that skill for each pull
-request promoted to Ready in this cleanup run. Post the secret gist preview as a
-pull request comment before completing the [Finish report](#finish-report).
-
-When the skill is absent, record `N/A` for the summary step and continue.
-
-## Promotion gates
-
-Run preflight work in isolated worktrees from the recorded parent SHA. Apply
-selected changes in the parent cleanup worktree after the owner selects the
-tested proposals.
-
-Promote the parent only after every actionable finding has an applied fix or an
-exact disposition. Record the remote parent Ready state and exact
-`parent_ready_sha`.
-
-Create the child from its intended pre-parent base and merge the exact
-`parent_ready_sha`. Prove that SHA is an ancestor of the child head with
-`git merge-base --is-ancestor <parent_ready_sha> <child_head>` and record exit
-code `0`.
-
-Reapply every relevant fix to the child. Rerun child tests, `e-simplify`, and
-`e-code-review` after the merge. Record the new child head and every validation
-result before promoting the child to Ready.
-
-When the repository ships `pr-summarizer`, run it on the child after child Ready
-promotion if the child is in scope for this cleanup run.
-
-Use `reference/task-seeds.md` and `reference/process-inventory.md` to record promotion evidence.
-
-## Finish report
-
-- Pull request or stack URLs.
-- `pr-refinement` outcome.
-- `source-command-sr-loop` passes, commits, and validation results.
-- Parent-ready and child-ready SHAs when a child is promoted.
-- `pr-summarizer` preview URLs and comment confirmation for each Ready pull request, or `N/A` when the repo does not ship the skill.
-- Remaining hard block, or `null`.
-
-## File index
-
-| Path | Purpose |
-|---|---|
-| `SKILL.md` | Hub for refinement, cleanup convergence, promotion gates, and reporting |
-| `reference/task-seeds.md` | Ordered session tasks for audits, delivery, validation, promotion, and summary |
-| `reference/process-inventory.md` | Process classes, evidence homes, and paired task checks |

package/.agents/skills/pr-cleanup/reference/process-inventory.md

@@ -1,17 +0,0 @@
-# PR cleanup process inventory
-
-| Step | Class | Home | Evidence | Paired test |
-|---|---|---|---|---|
-| Resolve PR and immutable parent SHA | deterministic | `task-seed:reference/task-seeds.md` | Target record and parent SHA | task-tool |
-| Create isolated preflight worktrees | deterministic | `task-seed:reference/task-seeds.md` | Worktree paths and base SHA | task-tool |
-| Run parallel preflight streams | borderline | `SKILL.md` and `task-seed:reference/task-seeds.md` | Stream reports and worker availability | task-tool |
-| Apply or disposition findings | judgment | `SKILL.md` and `task-seed:reference/task-seeds.md` | Changed diff or exact disposition | task-tool |
-| Validate parent head | deterministic | `task-seed:reference/task-seeds.md` | Scoped test and confirmation results | task-tool |
-| Promote parent and record exact Ready SHA | deterministic | `task-seed:reference/task-seeds.md` | Remote parent Ready state and SHA | task-tool |
-| Post parent PR summary gist comment | deterministic | `SKILL.md` and `task-seed:reference/task-seeds.md` | Preview URL, comment confirmation, or `N/A` | task-tool |
-| Merge exact parent SHA into child | deterministic | `task-seed:reference/task-seeds.md` | Child merge commit and parent SHA | task-tool |
-| Prove parent SHA ancestry | deterministic | `task-seed:reference/task-seeds.md` | `git merge-base --is-ancestor` exit code `0` | task-tool |
-| Reapply relevant fixes to child | judgment | `SKILL.md` and `task-seed:reference/task-seeds.md` | Child diff and reapplication record | task-tool |
-| Validate child and rerun confirmations | deterministic | `task-seed:reference/task-seeds.md` | New child-head tests, simplify, and review results | task-tool |
-| Promote child and report | deterministic | `task-seed:reference/task-seeds.md` | Child Ready state and finish report | task-tool |
-| Post child PR summary gist comment | deterministic | `SKILL.md` and `task-seed:reference/task-seeds.md` | Preview URL, comment confirmation, or `N/A` | task-tool |

package/.agents/skills/pr-cleanup/reference/task-seeds.md

@@ -1,22 +0,0 @@
-# PR cleanup task seeds
-
-Register each numbered item as one session task before work starts. Complete
-each task with evidence from the stated gate.
-
-1. Resolve the target PR, repository, parent head SHA, and child boundary.
-2. Create isolated preflight worktrees from the immutable parent preflight SHA.
-3. Run the shared-extraction, capability-name, `e-simplify`, and `e-code-review` preflight streams in parallel when available.
-4. Record each stream's worktree, base SHA, findings, proposed patch, and validation evidence.
-5. Keep parent merge, rebase, push, and Ready operations unavailable during preflight.
-6. Select, apply, validate, or exactly disposition every actionable preflight finding in the one-agent parent cleanup worktree.
-7. Run parent scoped tests and cleanup/review confirmation checks.
-8. Promote the parent to Ready and record the exact remote `parent_ready_sha`.
-9. When `.claude/skills/pr-summarizer/SKILL.md` exists in the repository under cleanup, run that skill on the parent and post the gist preview comment; otherwise record `N/A`.
-10. Create the child from its intended pre-parent base and merge the exact `parent_ready_sha`.
-11. Prove `parent_ready_sha` is an ancestor of the child head and record the exit code.
-12. Reapply every relevant preflight fix to the child.
-13. Run child scoped tests on the new child head.
-14. Rerun `e-simplify` and `e-code-review` on the new child head.
-15. Promote the child to Ready after all child evidence passes.
-16. When the repository ships `pr-summarizer` and the child is in scope, run it on the child and post the gist preview comment; otherwise record `N/A`.
-17. Write the finish report with SHAs, evidence, summary preview URLs, states, and hard blocks.

package/.agents/skills/pr-name-by-capability/SKILL.md

@@ -1,73 +0,0 @@
----
-name: pr-name-by-capability
-description: >-
-  Audit a GitHub PR for name-by-capability violations: driver/motive words on
-  reusable capability modules (queues and report routers may keep the driver
-  word), then apply capability-oriented renames by default. Triggers:
-  /pr-name-by-capability, name-by-capability audit, audit PR for naming, name
-  by capability, capability naming review, cert_fix rename check, driver word in
-  package name.
----
-
-# Name-by-Capability Audit
-
-Audit a GitHub PR against the **name-by-capability** rule. Inspect paths and title wording for driver or motive words on reusable capability code, and report violations with capability-oriented rename directions.
-
-## Mode routing
-
-First-match routing selects the mode before the audit steps:
-
-1. `preflight-proposal` runs `pr-name-by-capability preflight-proposal <pr_number> --base-sha <base_sha> --head-sha <head_sha> --worktree <isolated_worktree>` under the [shared preflight proposal contract](../_shared/pr-loop/preflight-proposal.md). Record each finding with violation or OK-driver classification.
-2. `audit-only` is report-only and ends after the findings report.
-3. A user-supplied rename or fix direction follows the existing normal fix workflow.
-4. Normal mode applies the suggested rename direction by default.
-
-## Gotchas
-
-- Name reusable libraries for the action they perform; a first cert caller still leaves the library capability-named. Score reuse potential alongside the PR title.
-- Correctly driver-named surfaces (`cert_fix_queue`, report routing/locate, unfixable patterns) are **OK** findings — list them under OK drivers.
-- Renames and new paths carry the most signal; when a PR expands an already-misnamed package, note the package name too.
-
-## When this applies
-
-- User invokes `/pr-name-by-capability <PR>` or asks to audit a PR for name-by-capability / capability naming.
-- PR adds or renames packages/modules, or frames a general shared operation with a driver/motive word (`cert_fix`, `cert_closeout`, `portal`, `export`, …).
-
-A missing PR number or URL returns exactly: `Give a GitHub PR number or URL to audit for name-by-capability.` Resolved targets follow **Mode routing**.
-
-## Constraints
-
-- Report findings plus a suggested rename direction for each violation.
-- Suggest renames for paths the PR touched; for an expanded offense, also name the package the PR grew.
-- Distinguish **violation** vs **OK driver** using `reference/offense-examples.md`.
-
-## Process
-
-Register every bullet from `reference/task-seeds.md` with `update_plan`. Mark each complete with evidence. Follow those seeds in order — do not restate them here.
-
-Load order for the rule: if `docs/agents/name-by-capability.md` exists, read it first; always keep `reference/rule-checklist.md` as the fallback when the doc is missing. On disagreement after the docs PR merges, prefer the repo doc and update the skill checklist in a follow-up.
-
-## Composition
-
-| Peer | Relationship |
-|------|----------------|
-| `pr-shared-extraction` | Layering and extraction (where code lives). Invoke when a path’s *role* is unclear; this skill only scores the *name*. |
-| `reviews` / PR review skills | May invoke this skill by name when naming is in scope |
-
-## File index
-
-| File | Purpose |
-|------|---------|
-| `SKILL.md` | Hub — gotchas, when-applies, process, composition |
-| `reference/rule-checklist.md` | Embedded rule + naming checklist (fallback when docs are missing) |
-| `reference/offense-examples.md` | Classifier + known OK drivers vs offenders |
-| `reference/fetch-commands.md` | Minimal `gh` fetch for PR naming surface |
-| `reference/report-template.md` | Compact report shape |
-| `reference/task-seeds.md` | Ordered task seeds for the audit run |
-| `../_shared/pr-loop/preflight-proposal.md` | Shared isolated proposal mode and evidence record |
-| `../_shared/pr-loop/preflight-proposal.contract.test.mjs` | Cross-skill mode and contract test |
-
-## Folder map
-
-- `SKILL.md` — hub
-- `reference/` — rule, examples, fetch recipe, report template, task seeds

package/.agents/skills/pr-name-by-capability/reference/fetch-commands.md

@@ -1,29 +0,0 @@
-# Fetch PR naming surface
-
-**Execute** these commands (pwsh). Replace `<N>` with the PR number. Parse the PR number from a URL (`…/pull/2153` → `2153`) when the user pastes a link.
-
-Prefer this recipe so the user’s worktree stays as-is. Do not `gh pr checkout` for naming audits.
-
-## Metadata
-
-```powershell
-gh pr view <N> --json number,title,body,baseRefName,headRefName,url
-```
-
-## Paths and renames
-
-One paginated files list covers added, modified, and renamed paths (`status`, `previous_filename`). Use `--paginate --slurp` per `gh-cli-conventions` (not built-in `--jq` across pages):
-
-```powershell
-gh api repos/:owner/:repo/pulls/<N>/files --paginate --slurp
-```
-
-When `:owner` / `:repo` are unknown, resolve them from `gh pr view <N> --json headRepository,url` or the PR URL, then re-run the files call.
-
-## What to keep from the fetch
-
-- PR title and first paragraph of body (scope wording)
-- Every added or renamed path under package roots and `.claude/skills/`
-- New Python module stems whose names carry driver words
-
-Scoring rules for those signals live in `reference/rule-checklist.md` — this file only fetches.

package/.agents/skills/pr-name-by-capability/reference/offense-examples.md

@@ -1,32 +0,0 @@
-# Offense examples (this repo)
-
-Classify by **role** — the reusable work the surface performs — not by matching a string list. For whether code belongs in `shared_utils` vs a workflow package, invoke `pr-shared-extraction`; this map only scores names.
-
-## Quick classifier
-
-```
-Queue, report parser, or routing table for one workflow only?
-  YES → driver name OK
-Shared artifact change a second workflow would call?
-  YES → capability name required (flag when a driver word is present)
-  Stays inside that workflow → OK driver; state why in the report
-```
-
-## OK — correctly driver-named
-
-| Surface | Why OK |
-|---------|--------|
-| `cert_fix_queue` | Rejected-theme queue drain — driver-only orchestration |
-| Cert report title → which ops to run | Driver / routing |
-| Cert report locate / folder routing | Driver-only path finding for cert artifacts |
-| Unfixable / unfixable-pattern tables | Cert-report classification artifacts |
-
-## Violations — driver word on reusable capability
-
-| Surface | Why violation | Suggested rename direction |
-|---------|---------------|----------------------------|
-| `cert_closeout` color engine (batch color UID rewrite) | General STP color batch shared across workflows | Capability name: STP color batch / color UID rewrite |
-| `cert_fix_*` modules that are general STP patches | Patch logic reusable outside cert closeout | Name the STP operation (asset replace, sound slot, …) |
-| `cert_fix_page_indicators` | Thematic page-indicator pair generation is a reusable asset op | Capability: page indicators / home indicator pair |
-| Portal/export-named helper that any workflow could call | Motive word on a reusable library | Name the shared action (sheet write, APK locate, …) |
-| PR titled or scoped with a driver word when the diff is a general capability | Motive framing on a library change | Retitle/scope to the capability; keep driver wording on the queue/driver PR |

package/.agents/skills/pr-name-by-capability/reference/report-template.md

@@ -1,29 +0,0 @@
-# Report template
-
-Emit this shape. Keep it compact.
-
-```markdown
-## Name-by-capability audit — PR #<N>
-
-**Verdict:** <clean | N violations, M OK drivers>
-
-### Violations
-| Finding | Path / title | Why | Suggested rename direction |
-|---------|--------------|-----|----------------------------|
-| Driver word on capability | `path/or/Title` | <one sentence> | <capability-oriented name family> |
-
-### OK drivers
-| Surface | Path / title | Why OK |
-|---------|--------------|--------|
-| Queue / routing / … | `path` | <one sentence> |
-
-### Notes
-- <optional: expanded offense in a package this PR grew, doc-only context, etc.>
-```
-
-### Rules for filling cells
-
-- **Finding** — short label for Violations rows only (`Driver word on capability`, `Motive framing in title`).
-- **Why** — cite the checklist item (reuse? action name? motive reserved for drivers?).
-- **Suggested rename direction** — a family of names (e.g. “STP color batch / color UID rewrite”).
-- When clean: **Verdict: clean** and omit Violation rows (keep the OK drivers / Notes sections if useful).

package/.agents/skills/pr-name-by-capability/reference/rule-checklist.md

@@ -1,38 +0,0 @@
-# Name-by-capability rule (skill-local)
-
-Prefer `docs/agents/name-by-capability.md` when that file exists in the worktree. This file is the fallback so audits still run when the docs PR has not merged yet.
-
-## Rule
-
-Name packages and modules for the **reusable work they perform**. Put a workflow or motive word (`cert`, `portal`, `export`, a product name, …) in the name only when that surface exists **solely** for that driver.
-
-1. **Capability name first.** Prefer names that say the operation (replace an asset slot, rewrite a color UID, send an email, locate a folder on disk, …).
-2. **Drivers stay at the edge.** Queue drains, report parsers, and other workflow-only shells may keep the workflow word. Libraries they call use capability names.
-3. **Orchestrators apply capabilities.** When several operations run on one artifact, one apply/orchestrator module owns stage → apply in order → result. Callers that know *why* map intents onto a registry of capability adapters.
-4. **Extend the capability, then wire the driver.** New use cases add or extend a capability adapter, then teach the driver which inputs map to it.
-
-## Checklist before accepting a new or renamed path
-
-A driver word in the name plus reuse by another workflow is a **violation**.
-
-1. Would another workflow reuse this? When yes, name the package/module for the shared action.
-2. Does the name state the action or artifact change?
-3. Is any motive word reserved for driver-only code?
-
-## Driver / motive words (non-exhaustive)
-
-Treat these as driver words when they appear in a package or module name for reusable library code:
-
-- `cert`, `cert_fix`, `cert_closeout`
-- `portal`, `export`, `submission`, `discount`
-- Product or campaign names used as a motive rather than an operation
-
-## What to score on a PR
-
-| Signal | Score when |
-|--------|------------|
-| New package/dir | Path segment carries a driver word |
-| Renamed package/module | Old or new name carries a driver word on capability code |
-| New module under an existing package | File/module stem carries a driver word for a general op |
-| PR title / scope | Frames a general shared capability change with a driver/motive word |
-| Unchanged misnamed package | Score when this PR **expands** the offense (new public API, new callers); otherwise list under Notes as out-of-diff context |

package/.agents/skills/pr-name-by-capability/reference/task-seeds.md

@@ -1,10 +0,0 @@
-# Task seeds
-
-Register each item with `update_plan` at audit start. Complete with evidence.
-
-1. Load the rule: `docs/agents/name-by-capability.md` when present, else `reference/rule-checklist.md` (always keep the skill checklist as fallback).
-2. Fetch PR title, body, and changed/renamed paths via `reference/fetch-commands.md`.
-3. Classify each naming signal as violation, OK driver, or skip (`reference/offense-examples.md` classifier first).
-4. Emit report from `reference/report-template.md`.
-5. After the report, continue with `SKILL.md` **Mode routing**.
-6. For `preflight-proposal`, complete the shared contract referenced by `SKILL.md` **Mode routing**.

package/.agents/skills/pr-plain-language-cleanup/SKILL.md

@@ -1,93 +0,0 @@
----
-name: pr-plain-language-cleanup
-disable-model-invocation: true
-description: Review every line of every changed PR file and trim words, names, docs, comments, and code while keeping real behavior.
----
-
-# PR plain-language cleanup
-
-## Balance compression and readability
-
-Shorten changed prose and make it easier to scan. Both matter.
-
-- **Word count:** record before and after counts and the delta. Count changed
-  prose only. Exclude code, frontmatter, and unchanged lines.
-- **Readability:** check sentence length, order, headings, jargon, and dense
-  blocks. Edit only when the result is easier to read.
-- **Diff churn:** count changed lines. Remove whitespace-only movement and
-  reflow-only churn.
-- **Meaning:** preserve behavior, APIs, data, test intent, conditions, warnings,
-  error detail, context, and public names.
-
-Reflow is allowed only when it materially improves scanability after a real trim.
-Report the four results separately.
-
-Review and trim this pull request:
-
-<PR LINK>
-
-If no link is given, find the clear PR in the current context. If none is clear, ask for one.
-
-Read the full PR, full diff, every changed file, changed tests and docs, and all repo rules that apply.
-
-Find the changed files. Send one capable agent to each file. If there are more files than agents, use small batches.
-
-Each agent owns one file. Read every line. Review code, tests, names, comments, docstrings, strings, notes, Markdown, and examples.
-
-Trim filler, repeats, vague words, needless code, needless names, dead notes, jargon, and steps that add no value. Use few words and small words.
-
-Keep real behavior, checks, data, error detail, context, tests, and public names unless a safe full rename is clear. Do not change behavior for style. Do not shorten a name when it loses meaning.
-
-Follow the repo rules:
-
-- Name modules for their capability.
-- Name functions for their action.
-- Name variables for the value they hold.
-- Name classes for what they are.
-- Use capability names for shared code.
-- Keep workflow words on workflow code.
-- Keep comments and docs true to the code.
-- Use real data and real behavior in tests.
-
-## Hard limits
-
-This is a non-material, text-only trim.
-
-DO NOT:
-
-- run `e-simplify`, `e-code-review`, or `source-command-sr-loop`;
-- run `pr-cleanup`, `pr-refinement`, `pr-small-cl`, placement audits, extraction audits, or capability-name audits;
-- run review loops, advisor gates, commit gates, promotion gates, or any other extra cleanup;
-- hunt bugs, change behavior, change APIs, change data, add features, or split the PR;
-- turn a text trim into a code fix.
-
-Run only the smallest checks needed for the files changed. Keep the repo rules. If a material issue appears, report it and stop.
-
-If a change needs another file, report it to the main agent. The main agent owns cross-file changes.
-
-After the file agents finish, apply safe edits, fix cross-file names and links, search for old names and wording, and read every changed file and the full diff again.
-
-Run only the smallest relevant check for the files changed. Fix failures. Check the final diff against the repo rules.
-
-Do not invent behavior, tests, or benefits. Stop and ask only when a change could alter behavior, break a public API, or exceed the PR scope.
-
-Commit and push the changes to the PR branch.
-
-Return:
-
-## What changed
-
-Short list of the main trims.
-
-## Files checked
-
-Every file reviewed.
-
-## Verification
-
-Checks run and results.
-
-## Remaining items
-
-Only items that need a user choice.
-

package/.agents/skills/pr-refinement/SKILL.md

@@ -1,31 +0,0 @@
----
-name: pr-refinement
-description: Coordinate shared-extraction and capability-naming audits, then update an existing pull request or publish a focused GitHub pull-request stack. Use when a user asks to turn a PR architecture audit into extracted shared code and small pull requests.
----
-
-# PR Refinement
-
-Turn audit findings into a focused existing pull-request update or a dependency-ordered GitHub pull-request stack.
-
-## Peer skills
-
-1. Run `pr-shared-extraction` for reusable-code placement, canonical shared homes, and extraction findings.
-2. Run `pr-name-by-capability` for module, symbol, path, branch, and pull-request naming findings.
-3. Use `pr-small-cl` to divide approved fixes into coherent, independently reviewable pull requests.
-
-The peer skills own their audit rules and finding priorities.
-
-## Workflow
-
-1. Run both audits in parallel against the same pull request. Keep their findings, locations, priorities, destinations, API directions, and naming directions.
-2. Build one change map. Group findings by shared capability and dependency.
-3. Use `pr-small-cl` to choose the delivery shape. Update the existing pull request when it remains one coherent, reviewable outcome.
-4. Create a replacement pull-request stack when the focused-change review identifies independent increments. Give each pull request one coherent outcome, related tests, a clear verification boundary, and a capability-oriented branch and title. Close the original pull request as superseded and link the replacement stack.
-5. Implement the existing pull request or the replacement stack in dependency order. Earlier pull requests create stable shared foundations. Later pull requests migrate consumers and remove replaced code.
-6. Run scoped production-path tests for each pull request. Record commands and results.
-7. Commit and push each branch. Update the existing pull request, or open draft replacement pull requests in stack order and set every child pull request base to its parent.
-8. Include scope, verification, risks, and stack dependencies in every pull-request body.
-
-## Completion
-
-Deliver a focused, tested, pushed existing pull request, or a replacement stack with explicit parent-child links. Keep merge authority with the user.