Skip to content

docs(skill): topology-base PRs and guideline PR count for CyClaw-Optimize - #811

Merged
cgfixit merged 1 commit into
mainfrom
grok/cyclaw-optimize-topology-wording
Aug 6, 2026
Merged

docs(skill): topology-base PRs and guideline PR count for CyClaw-Optimize#811
cgfixit merged 1 commit into
mainfrom
grok/cyclaw-optimize-topology-wording

Conversation

@cgfixit

@cgfixit cgfixit commented Aug 5, 2026

Copy link
Copy Markdown
Owner

What

Wording-only update to CyClaw-Optimize (skill + slash command mirrors):

  1. After Step 3.5 — stop treating every chunk as “cut from origin/main / base=main.” Step 4, branch permissions, and gotchas now use the topology base from 3.5 (independent → main; stacked → parent branch; consolidated → owner branch). MCP examples cover both independent and stacked base.
  2. Session merge order — prefer merging this run’s drafts lowest PR number → highest, with stack parents always before children; refresh children after a parent lands (squash-aware).
  3. PR count is a guideline — about 1–4 when several fixes earn their keep; one solid PR beats four thin ones; zero is fine; do not invent work to fill a slot. Softened Step 1/3/example “~5” language accordingly.

No runtime code, bootstrap, or invariant changes.

Why / benefit

Reduces the “five peers from main + half-stack” steering that produced dirty stacked children (e.g. #803 after #802) while keeping Step 3.5’s consolidate/stack mechanics.

Risk to monitor

  • Agents may over-stack when independent non-adjacent edits would merge cleanly — still allowed when Step 3.5 finds no shared-file dependency.
  • Lowest-number merge order can disagree with parent-first if open order was wrong; parent-first is stated to win.

Merge topology

  • independent (docs-only skill/command wording)
  • no shared-file stack with open harness PRs

Test plan

  • Diff limited to .claude/skills/CyClaw-Optimize/SKILL.md and .claude/commands/CyClaw-Optimize.md
  • Skim Step 4 table + merge-order paragraphs for clarity
  • Squash-merge with a clean final message when ready

…mize

Reword post-3.5 steps so branches/PRs use the Step 3.5 topology base
(not a blanket origin/main default), document session merge order as
lowest PR number to highest with parent-before-child when stacked, and
treat focused PR count as a guideline rather than a quota.
@cgfixit
cgfixit marked this pull request as ready for review August 6, 2026 02:45
@cgfixit
cgfixit merged commit 8033c3e into main Aug 6, 2026
42 checks passed
@cgfixit
cgfixit deleted the grok/cyclaw-optimize-topology-wording branch August 6, 2026 02:45

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cc2c05add9

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".

Comment on lines +215 to +216
lands on `main` (often via squash), refresh each open child onto current
`origin/main` (or retarget base to `main` after rebase) before merging the next.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Use an onto rebase after squash-merging a parent

When a stacked parent contains multiple commits and is squash-merged, an ordinary git rebase origin/main on the child replays the parent's original commits from the old merge base; this can conflict or reintroduce parent changes. The new squash-aware instruction says only to “refresh” or rebase, so direct the agent to replay only child commits with git rebase --onto origin/main <parent-tip> <child> (or an equivalent child-only operation) before retargeting the PR.

Useful? React with 👍 / 👎.

Comment on lines +18 to +20
groups findings into a **small set of reviewable PR-sized chunks** (about 1–4
when several independent fixes earn their keep — **a guideline, not a quota**;
do not invent low-value PRs to hit a number), and opens one focused pull request

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Propagate the non-quota policy to the other optimizer docs

Changing the canonical Claude workflow to a 1–4 guideline leaves CLAUDE.md:424-428 describing the Kimi port as producing ~5 PRs and .codex/skills/cyclaw-optimize/SKILL.md:19-22 retaining the same quota language. Invocations through those documented agent paths can therefore continue the behavior this commit intends to stop; update the derived docs/ports or explicitly document why they diverge.

AGENTS.md reference: AGENTS.md:L78-L82

Useful? React with 👍 / 👎.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant