Skip to content

Add a README, and gate pull requests on CI again - #381

Merged
ramesh130 merged 1 commit into
mainfrom
readme-and-pr-triggered-ci
Sep 21, 2026
Merged

ramesh130 merged 1 commit into
mainfrom
readme-and-pr-triggered-ci

Conversation

@ramesh130

Copy link
Copy Markdown
Owner

Two things a repository needs before anyone outside it can use or change it,
neither of which this one had.

A front door. Everything a newcomer needs was already written and none of
it was addressed to them: CLAUDE.md orients an agent mid-change, PRD.md is
the roadmap the issues are cut from, and docs/ answers a reader who has
already decided to be here. README.md is for somebody who has not.

It leads with what the project does not promise rather than with the feature
list, because the feature list is by far the longer of the two: nothing is
published or resolvable by anybody, 0.x may break on a minor, most phases'
device exit criteria are open, the benchmark's Phase 3 verdict is a committed
finding rather than a win, superplayer-ui is an empty placeholder, and
superplayer-moq is local-only with #369 unanswered. Read in the other order
the same repository would mislead, which is the only real risk in publishing
it. Every count and link is checked against the tree rather than recalled --
eighteen decision records, ADR-0001 through ADR-0018, since 0000 is the
template.

docs/README.md gains it as a seventh root document, with what makes it
different from the other six.

Per-change CI. The dispatch-only trigger was a reasonable arrangement for
one person on a private repository and an unreasonable thing to ask of a
contributor: it makes "it passed on my machine" the only evidence, in both
directions. The budget that justified it lapses with privacy, since standard
runners are free for public repositories.

pull_request is the load-bearing half -- a pull request from a fork is not a
push to this repository, so push alone never sees one and a fork's change
would arrive ungated. push is restricted to main, where it checks the
result of a merge rather than the branch that was merged: a rebase replays
commits onto a head none of them was tested against. Leaving it unrestricted
would run every topic branch twice for one answer, and a branch with no pull
request open is still reachable through workflow_dispatch.

No library code changes. spotlessCheck and verifyLicenseHeader pass, the
workflow parses and its triggers read back as intended, and every relative
link in both READMEs resolves.

Verification

No library code changes. spotlessCheck and verifyLicenseHeader pass, the workflow parses and its triggers read back as intended, and every relative link in both READMEs resolves.

Nothing gated this merge. GitHub Actions cannot run on this account at present — a dispatched run was refused with "recent account payments have failed or your spending limit needs to be increased" (#380). Local verification is the only evidence.

🤖 Generated with Claude Code

https://claude.ai/code/session_01HeTWwmK1KcFMrfSAre23GS

Two things a repository needs before anyone outside it can use or change it,
neither of which this one had.

**A front door.** Everything a newcomer needs was already written and none of
it was addressed to them: `CLAUDE.md` orients an agent mid-change, `PRD.md` is
the roadmap the issues are cut from, and `docs/` answers a reader who has
already decided to be here. `README.md` is for somebody who has not.

It leads with what the project does not promise rather than with the feature
list, because the feature list is by far the longer of the two: nothing is
published or resolvable by anybody, `0.x` may break on a minor, most phases'
device exit criteria are open, the benchmark's Phase 3 verdict is a committed
finding rather than a win, `superplayer-ui` is an empty placeholder, and
`superplayer-moq` is local-only with #369 unanswered. Read in the other order
the same repository would mislead, which is the only real risk in publishing
it. Every count and link is checked against the tree rather than recalled --
eighteen decision records, `ADR-0001` through `ADR-0018`, since `0000` is the
template.

`docs/README.md` gains it as a seventh root document, with what makes it
different from the other six.

**Per-change CI.** The dispatch-only trigger was a reasonable arrangement for
one person on a private repository and an unreasonable thing to ask of a
contributor: it makes "it passed on my machine" the only evidence, in both
directions. The budget that justified it lapses with privacy, since standard
runners are free for public repositories.

`pull_request` is the load-bearing half -- a pull request from a fork is not a
push to this repository, so `push` alone never sees one and a fork's change
would arrive ungated. `push` is restricted to `main`, where it checks the
result of a merge rather than the branch that was merged: a rebase replays
commits onto a head none of them was tested against. Leaving it unrestricted
would run every topic branch twice for one answer, and a branch with no pull
request open is still reachable through `workflow_dispatch`.

No library code changes. `spotlessCheck` and `verifyLicenseHeader` pass, the
workflow parses and its triggers read back as intended, and every relative
link in both READMEs resolves.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HeTWwmK1KcFMrfSAre23GS
@ramesh130
ramesh130 merged commit 7de82a2 into main Sep 21, 2026
0 of 2 checks passed
@ramesh130
ramesh130 deleted the readme-and-pr-triggered-ci branch September 21, 2026 04:10
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