agentsclimarketplace

Evergreen comments

Skill cardstack/boxel/.claude/skills/evergreen-comments

Write code comments, PR descriptions, PR review replies, and other reader-facing prose so they describe the code as it is now — never the journey of how it got there, and never a private ticket/PR reference. Use whenever writing or editing a code comment, a PR title/description/comment, or restating a tracker ticket. Triggers on adding/editing comments, opening or updating a PR, or replying to review feedback.From its SKILL.md

Install
npx -y skills add cardstack/boxel --skill evergreen-comments

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

6.3 KB, as published. Nobody here has run it

Evergreen Comments & Prose

Reader-facing prose describes the current contract, stated as timeless fact. It does not narrate how the author arrived, and it does not reference private trackers or sibling changes. Delivery state and iteration history rot the moment PRs renumber, tickets close, branches reland, or the work gets restructured — and the reader years later cares about the invariant the code upholds, not the path to it. Archaeology is git log / git blame territory.

This codebase is open source; the issue tracker is private. A ticket ID in a comment is opaque to almost everyone who will ever read it.

Scope. These rules govern prose you write or introduce: code comments, skill files, PR descriptions, PR review replies, and tracker restatements. Two things are deliberately outside it: commit messages (see Where the journey legitimately goes) and pre-existing prose you're merely editing around — you need not clean up rot you didn't write.

The two rules

1. Present the code as it is — not the journey

Strip anything about how the author got here. Within a single PR especially, the journey is worthless to a future reader.

Always cut:

  • "An earlier revision / earlier attempt did X", "First I tried A, then B, then C"
  • "Originally we did X but switched to Y", "the first cut was X, redesigned to Y" — this describes a path not taken
  • "I reproduced locally, narrowed to Y, reverted Z", "confirmed by reverting…", "diagnostic logs pinpointed…"
  • Iteration timelines, SHA-by-SHA reports, "earlier this broke six tests"
  • Temporal / relative-time language: "until now", "used to", "previously", "the old behavior was", "now we", "recently", "as of today". A reader has no anchor for when "now" is — the comment reads as if something just changed even when it changed years ago. State the contract timelessly.

2. No private-tracker or sibling-change references

Never write a real one of these into code comments, skill files, PR descriptions, or review replies. (Clearly-fake illustrative placeholders — like the anti-pattern examples in this very skill — are the obvious exception; you can't teach the rule without showing what it forbids.)

  • Ticket IDs — a real CS-… identifier; the private tracker is meaningless and opaque to outside readers
  • PR numbers / PR-letter labels — #4863, "PR A", "PR B", "stacked PR N", "in this PR we…"
  • Sibling-change references — "same pattern the prior PR added", "see also that other PR's comments"

Name the mechanism, not the change:

  • ❌ "PR A's PagePool fix makes the on-demand path safe."
  • ✅ "The PagePool's tab-materialization for module/command callers makes the on-demand path safe."

Example: same content, evergreen vs. rotty

Rotty (don't write) — encodes delivery state, references a private ticket and a sibling PR, and uses "no longer":

// CS-XXXXX PR 3: the /_atomic endpoint no longer awaits indexing.
// We KEEP /_atomic (rather than switching to parallel +source POSTs)
// ...same pattern PR 2 added for the +source surface.

Evergreen (write this) — same technical content, survives the next refactor:

// /_atomic returns once writes are durable, not once they are indexed.
// Callers needing indexed state must drain via realm.incrementalIndexing()
// (server-side) or wait on the matrix 'index' incremental event (client-side).
// Mixed module+instance batches are serialized: the intermediate flush in
// _batchWrite is always awaited, so an instance's serialization sees its
// module already indexed.

PR descriptions and review replies

A PR description carries the current shape only: what the code does now, why this shape is right (constraints/trade-offs as live invariants, not "we tried X and it failed"), what's in scope vs. deferred, and the test plan. When a PR pivots, the description gets rewritten to the new shape — not a changelog of the pivot.

A status comment after a re-push is a single sentence pointing at the description — nothing else. The iteration history goes in the private tracker or commit messages, or nowhere.

A review reply that reports a finding addressed always names the commit: "fixed in abc1234", then the evergreen statement of the current contract. Commit SHAs are the one pointer that belongs here — they are repo-native, exactly what git blame/git log speak, so they never rot the way ticket IDs and PR-number cross-references do, and they tell the reviewer precisely which commit to re-review. Ban the tracker refs, keep the SHA.

Restating a tracker ticket

When a ticket's original framing predates data you now have, rewrite it as a clean statement of the current problem → live measurements → path forward. Not "originally we thought X, but now we know Y", not "PRs landed since last update". No chronology.

Where the journey legitimately goes

  • Commit messages — exempt from the no-journey and temporal rules: one substantive change each, and the chronology / "why now" belongs here.
  • The private tracker — staff-only, full detail welcome.
  • git blame / git log — the canonical history. Nothing in a comment competes with it.

Self-check before publishing

Read it back. If a sentence starts with "Earlier", "Before", "Originally", "First I", "I tried", "I reproduced", "Confirmed by", "After reverting", "until now", "used to", "previously", or "now we" — that's journey or temporal rot. Cut it. If it names a ticket ID, PR number, or PR letter — cut that too. If what remains explains what the code does now and why, ship. If it explains how you arrived, cut more.

As stated in Scope above: pre-existing rotty references in code you're editing can stay — the rule governs prose you write, not cleanup of what's already there.

Keep looking

Skills are one crate of 613,757. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.