Skip to content

docs(adr): Drop ADR-0022's hardcoded harness count - #778

Merged
omesser merged 2 commits into
mainfrom
fix/776-777-census-and-adr
Sep 16, 2026
Merged

omesser merged 2 commits into
mainfrom
fix/776-777-census-and-adr

Conversation

@omesser

@omesser omesser commented Sep 16, 2026 •

Copy link
Copy Markdown
Owner

Why

ADR-0022 said the launch table "names five harnesses (claude, codex, grok, hermes, opencode)". HARNESS_PRESETS holds seven. pi and then cursor-agent (#762) landed without reconciling the record, so it has gone stale twice in a row, silently.

The count was never the decision. ADR-0022 decided focused list plus escape hatch over enumerate everything, and that is true at five, seven, or twelve.

What changed

One paragraph in docs/adr/0022-acp-client-over-official-sdk-and-named-harnesses.md. The count and the enumeration are gone; the policy and a pointer to HARNESS_PRESETS replace them.

Deliberately not done:

  • No code. HARNESS_PRESETS is untouched.
  • The symbol, not the path. docs/agents/docs.md rules file paths and constant names out of an ADR because mechanism is fluid. A symbol name survives the refactor a path does not, so the ADR names HARNESS_PRESETS and leaves src-tauri/src/settings/form.rs to grep. Naming the path would re-arm the same rot this PR is removing.
  • The escape hatch and the smoked-turn rule stay. Both are load-bearing and both are still correct.
  • ADR-0017 left alone. It is superseded by this one, and a superseded record is history.
  • No other copy to fix. README.md:157-176 and docs/DEVELOPMENT.md:247 both already list all seven. The ADR was the only stale count.

#776 is not in this diff. It needs scripts/comment-census.py, which #758 is still creating, so the fix went there as a review comment rather than as a second PR racing the same new file: #758 (comment)

How to verify

node --test tests/*.test.js at a9924690, rebased onto 39fd012c: 212 pass, 0 fail. pre-commit run --files docs/adr/0022-acp-client-over-official-sdk-and-named-harnesses.md passes, including codespell and the ADR-number hook.

grep -rn 'names five' docs/ README.md returns nothing.

Notes

The open question is now answered, and f14ce9e2 writes it down. #762 raised it and left it open: does a verified handshake earn a named row, or does a smoked turn? @omesser settled it. Verified means smoked, not handshaked, and Copilot CLI has not been verified.

ADR-0022 already stated the rule correctly one paragraph up, "Harnesses earn verified standing once a turn has been smoked against them, fresh and resumed." The ambiguity was one loose word downstream, where the same record said a harness earns a named row "once verified". That reads as either answer. It now says which:

-through the custom command and earn a named row once verified. Google is not
+through the custom command and earn a named row once a turn has been smoked
+against them. Protocol compatibility alone does not earn the row: Copilot CLI
+has it and is still not named.

Nothing else changes. The README and docs/DEVELOPMENT.md already match this rule, and the Cursor row #762 added was smoked to end_turn before it merged, so it keeps its standing.

Closes #777

🤖 Generated with Claude Code

— Claude Code agent, on @omesser's behalf.

omesser and others added 2 commits September 16, 2026 17:59
The count went stale twice. Point at HARNESS_PRESETS instead.

Closes #777

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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.

docs(adr): ADR-0022 says the launch table names five harnesses; it names seven

1 participant