Repository navigation
docs: fix step references stale after the 63-unit re-split - #22
Conversation
The s217 re-split renumbered ci/PROMPT.md to 63 steps in seven phases but
left three cross-references pointing at earlier numbering schemes:
- ci/feedback/README.md called the skeleton feedback PR "step 23"; it is
step 49. Step 23 is now the unit-test layer, so the link read as
plausible while pointing at the wrong step.
- README.md still described "eleven checkpoints in six phases, one PR
each" — two schemes out of date, and "one PR each" contradicts the
decoupling of grind units from the 14 fixed PR boundaries. It also
claimed two phases carry no checkpoint, written before phase 7.
- PROMPT.md own stop-condition prose kept the pre-split arithmetic
("the remaining 59", "52 of 54 steps").
Range references verified sequential and in-range; no inverted ranges
survive from the renumbering pass.
WalkthroughChangesAdoption workflow documentation
Possibly related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
✨ Simplify code
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: 99c9258b-88e5-44f6-a72d-b3b9472f9f7a
📒 Files selected for processing (3)
README.mdci/PROMPT.mdci/feedback/README.md
| 63 steps in seven phases, covering: inventory, the decision seam, the `ci/` | ||
| layout, runner identity, workflows + badges + a single entry point, the four | ||
| test layers, fuzzing, caching + the linter gate, lanes, self-hosted exposure, | ||
| and a depth pass asking whether any of the now-green gates would actually catch | ||
| anything. A step is a **grind unit, not a PR** — it is sized to be handed to a | ||
| cheap model with the file and nothing else, and steps land in groups across 14 | ||
| fixed PR boundaries that must not be merged partially. The last phase is the | ||
| aftermath: the checks worth running once the adoption is green. Steps 1–6 work | ||
| out how much of the standard the target already has, so you do not need to know | ||
| in advance whether this is a first adoption or a top-up. |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win
Align the PR-granularity rule with the workflow table.
README.md now states that steps land in groups across 14 fixed PR boundaries. ci/PROMPT.md Lines 39–58 defines those groups. However, ci/PROMPT.md Lines 82–84 still says “One PR per step that lands code” and requires the previous step to merge first. An adopter cannot follow both instructions. Update the standing constraint to describe the fixed PR groups, including the 12+ depth-pass exception.
| It is a **merge into whatever CI the target already has**, not a greenfield | ||
| install — the target's tests, thresholds and extra gates survive; the layout, | ||
| ordering, naming and entry points become the standard's. A final section covers | ||
| the inverse job once a module is fully adopted: forwarding one later skeleton | ||
| change into it. | ||
|
|
||
| **Settle the runner question first** (checkpoint 2 of the prompt). These | ||
| **Settle the runner question first** (steps 13–15 of the prompt). These |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win
Align the runner prerequisite step number.
README.md points adopters to steps 13–15. The phase and PR tables in ci/PROMPT.md also assign runner identity to steps 13–15. However, ci/PROMPT.md Line 77 still says “Step 7 settles it.” Step 7 is the decision-seam phase. Update the stale reference so the adoption sequence has one runner prerequisite.
Proposed documentation fix
- Step 7 settles it before any workflow is ported.
+ Steps 13–15 settle it before any workflow is ported.
1 participant
TL;DR
ci/PROMPT.mdwas renumbered to 63 steps in seven phases, but three documents that point into it still use older numbering. A reader following any of them lands on the wrong step, or on a step that no longer exists.What was stale
ci/feedback/README.mdcalls the skeleton feedback PR "step 23". It is step 49; step 23 is now the unit-test layer. The link stayed plausible while pointing somewhere else entirely, which is the worst kind of stale cross-reference.README.mddescribed "eleven checkpoints in six phases, one PR each" in two places. That is two numbering schemes out of date, and "one PR each" now contradicts the split: a step is a grind unit, and steps land in groups across 14 fixed PR boundaries. It also claimed two phases carry no checkpoint of their own, written before phase 7 existed.ci/PROMPT.mdkept pre-split arithmetic in its own stop-condition prose: "carry on with the remaining 59" and "finishes 52 of 54 steps", against a 63-step file.Testing
Docs only, no code paths touched. Verified by grep:
checkpoint,eleven,six phases,25 stepsor56 stepssurvives in any tracked.mdsteps N–Mrange inPROMPT.mdis sequential and in range; no inverted ranges survive from the renumbering pass## 49 — Hand the findings back to the skeleton, the step that writesci/feedback/<target>-<YYYY-MM-DD>.mdmarkdownlintreports MD013/MD025/MD060 acrossPROMPT.md, all pre-existing and none on the changed lines. Left alone rather than mixed into a reference fix.