@AlekVectis Hidden check catching what the suite already passed is the lesson. Gate the ticket before the run starts: if done can't name a check outside the agent's write path, don't start. "All tests pass" alone never proved the bug was gone.
@sukin_s That failure mode is exactly why green isn't done. If the same agent wrote the harness, it can encode the implementation. Keep tests the agent can't loosen locked, and treat deleted or skipped tests as a fail after the run — not a success story.
@themuqsit Check the diff, not the status — yes. Pair that with a path-scope gate after the run: if the ticket named one package and the diff touched four, reject even when 239 tests are green. Status lied; the diff can't.
@Rana_kamran43 This is the trap. "All tests pass" is a claim the agent can satisfy by editing the suite. Write done as a check: refuse vague tickets before spend, then compare the diff to allowed paths and flag deleted or skipped tests. ContextMux. See how it works: https://t.co/DVGglKoQja
Keep the rules you have.
ContextMux's init imports your CLAUDE.md, AGENTS.md, copilot-instructions and .cursor/rules into .ctxmux/, then compiles them back out.
No key. Undo: git checkout . && git clean -fd
See how it works → https://t.co/DVGglKpo8I
@benbarbersmith Teaching agents in AGENTS.md is the right PR. For monorepos the next step is path scope: one list that becomes Cursor globs and Copilot applyTo for the same folders, so a package rule doesn't leak into every other app.
@Hunchopreneur Same idea, every tool is right. The catch is keeping those four files in lockstep after the first export. Write the rule once, emit Cursor / Copilot / Claude copies, and fail CI when a hand-edited copy drifts.
@Civil3DAngel One line between Claude and Codex is clean. Once you also need folder rules, formats split again: Cursor wants globs, Copilot wants applyTo, Claude wants paths in its rules. Same list, three files — or generate those three from one source so they can't drift.
@shtse8 Solid map. One gap teams still hit: path-scoped rules get hand-copied into .mdc globs, applyTo, and Claude paths, then drift. Keep one list, emit each format, use sync --explain for what each tool loses. ContextMux. See how it works: https://t.co/DVGglKoQja
Today ctxmux writes scoped rules for Claude as labelled sections in CLAUDE.md.
Native where ctxmux can map it, flagged where it can't. No silent loss.
See how it works → https://t.co/DVGglKoQja
Is hand-copying scoped rules the accepted tax?
Cursor wants .mdc globs. Copilot wants applyTo. Copy the same folder rule into both, and the two drift apart by the next refactor. 🧵
Not every target maps 1:1, so ctxmux sync --explain tells you what each one loses:
Copilot: full fidelity
Cursor: nothing lost functionally
Codex: cross-cutting globs become advisory
Claude: path scoping is advisory
@akashsairaja Under a page here too. The habit that keeps mine honest: every line has to still be true today, and each rule lives in one place only. Once it's copied into AGENTS.md or Cursor rules, the copy is the one that goes stale first.
@RobertLibsansky Two kinds of outdated. Wrong facts need a review. Copies that disagree (CLAUDE.md vs AGENTS.md) are mechanical: generate them from one source and fail CI on drift. ContextMux's check does that part. See how it works: https://t.co/DVGglKoQja
@kirillk_web3 The quiet half of this: an old rule doesn't error. The model keeps obeying it, and it looks like the model got worse. Delete old lines in one place, not three. If CLAUDE.md, AGENTS.md and Cursor rules each hold a copy, the stale one survives.
@adocomplete Worth knowing as that file grows: Claude follows it whether it's still true or not. A stale line never errors, it just gets obeyed. If the same rules also sit in AGENTS.md or Cursor rules, generate those copies from one source and fail CI when one drifts.
A stale rule doesn't throw. The agent quietly follows it, and you blame the model.
ContextMux catches it: advise flags duplicate or contradicting rules and dead paths, and check --strict fails CI when a copy drifts from .ctxmux/.
See how it works → https://t.co/DVGglKoQja
@calvinellis75@agents Steal away 😄 The @-import is a clean setup for Claude and Codex. Scoped rules are where it splits again: Cursor wants .mdc globs, Copilot wants applyTo, Claude wants .claude/rules, so folder rules end up written three times. Worth covering those in the drift check too.