Tae skills effnext skip and paperwork

[jjSDET]

tae: correct the conversion loop's paperwork steps and document effnext's skip flow

Three documentation changes to the tae skills, all from running the conversion loop end to end on a real stack. Two of them correct instructions that were actively wrong; the third documents new tooling.

Depends on the companion testops-tools PR (tae-conversion-toolchain-fixes) — it adds the effnext flags described here and the effverify behaviour this corrects. Land that one first so these skills don't reference flags that don't exist yet.

1. Paperwork the loop was getting wrong (efficiency-conversion-loop)

Three rules, each of which cost rework on the most recent stack:

• Conversion bugs must block the tracking meta (2030727), set at create time. The meta tracks the campaign through its depends_on list, so a conversion bug that never gets linked is invisible to anyone reading it for status. Two bugs on the last stack were caught unlinked during a pre-submit audit. Scope matters too: tooling, docs and harness bugs stay off the meta, since it is specifically about migrating and removing legacy tests. Includes the backfill form, and a warning never to PUT a bare list to a meta's depends_on — Bugzilla treats that as replace, which would silently drop every other bug it tracks.
• @Converted goes in the same commit as the conversion, gated on green locally rather than on landing. The burndown keys off that marker, so a conversion landing without it reads as unconverted; annotating afterwards needs a second bug and a second review. On the last stack this was missed on two conversions and required a mid-stack rewrite to fix properly. The commit's paths list must include the annotated legacy file.
• Jira goes through tools/jiratool.py, which works headless, rather than the Atlassian connector, which is absent in headless and cron runs. Notes that create defaults to Story with no labels and does not self-assign, so all three have to be passed explicitly.

2. effnext's skip flow (efficiency-test-authoring, gate 0)

Whoever picks up the queue may not be in a position to take the next candidate — too complex for them, blocked on a harness gap, or deliberately deferred by whoever is sequencing the campaign. There was no way to say so, so the decision lived in someone's head and the next caller got the same pick.

Gate 0 and the tool table now describe --skip Class.method --reason "…", --unskip, and --skips, including the property that makes them safe to use casually: a skip never marks a test converted.

The same entry documents the new in-tree filter. The done-ledger lags the tree — it proposed a test that was already converted and committed, and 11 of its top 30 candidates were already in-tree — so effnext now drops candidates whose method already exists in the efficiency tests package. Worth knowing that effscaffold's already_converted does not cover this: it matches files, not methods, which is how the stale pick got past gate 1.

3. Two traps at the assertion gate (efficiency-test-authoring)

From bug 2060405, both of which cost a long debugging session:

• A disabled Compose button still accepts the click gesture and silently skips onClick, so "clicked" in the report means the gesture was delivered, not that the app acted. The failure then surfaces wherever the effect was expected — in that case 25 seconds and several steps later.
• An enabled-check against a COMPOSE_BY_TEXT selector is a no-op, because it resolves the text node inside the button, which reports enabled while the button is disabled. Use COMPOSE_BY_TEXT_MERGED for anything you act on.

Plus the rule that falls out of both: prefer a positive assertion over waiting for something to disappear, since absence cannot distinguish "it worked" from "the click was dropped".

4. effverify's scope was documented as its own bug

The tool table said the done-gate is "scoped to the last run". That was the defect, not the contract: effloop emits one run block per test after the class run, so reading only the last block reported passing tests as not-run, genuinely failed tests as failed_total: 0, and retried: false across a run containing eight of them.

effverify now aggregates every block and reports a per-test status including retry-pass. Documenting the broken behaviour as intended is worse than not documenting it — an agent would trust a verdict the tool no longer produces, and have no reason to question a green that came from the wrong block.

Notes

Documentation only; no plugin code changes. The rules here are not proposals — each one is the corrected version of something that went wrong on the four-commit Fenix stack these skills drove, which shipped two conversions, two harness fixes, and a full-suite sweep of 42 efficiency classes / 137 tests.

[Copilot]

Pull request overview

Updates the tae skill documentation based on an end-to-end run of the ui/efficiency conversion workflow, correcting previously incorrect instructions and documenting new/updated tool behaviors.

Changes:

• Document effnext skip/unskip flow and in-tree filtering in the authoring gate 0 instructions.
• Clarify assertion-gate Compose pitfalls and reinforce @Converted timing/placement requirements.
• Update conversion-loop “paperwork” steps to require blocking the tracking meta, using headless Jira tooling, and describing effverify’s aggregated-run behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
plugins/tae/skills/efficiency-test-authoring/SKILL.md	Documents effnext skip behavior, adds assertion-gate gotchas, and tightens @Converted workflow guidance.
plugins/tae/skills/efficiency-conversion-loop/SKILL.md	Corrects conversion-loop steps (meta blocking, @Converted in-commit) and switches Jira instructions toward tools/jiratool.py.

Suppressed comments (1)

plugins/tae/skills/efficiency-conversion-loop/SKILL.md:138

• This section says to use tools/jiratool.py, but the command examples invoke python3 jiratool.py, which will fail unless the reader happens to be in the tools/ directory. Use an explicit path (or state the required working directory) so the commands are copy/pasteable.

python3 jiratool.py create '<summary>' --file body.txt --parent MTE-5731 --issuetype Sub-task --label conversion
python3 jiratool.py create '<summary>' --file body.txt --parent MTE-5715 --issuetype Sub-task --label enablement
python3 jiratool.py link <enablement-key> Relates <conversion-key>
python3 jiratool.py assign <key> --me        # create does NOT self-assign; do this explicitly

</details>



---

💡 <a href="/mozilla/firefox-aidev-plugins/new/main?filename=.github/instructions/*.instructions.md" class="Link--inTextBlock" target="_blank" rel="noopener noreferrer">Add Copilot custom instructions</a> for smarter, more guided reviews. <a href="https://docs.github.com/en/copilot/customizing-copilot/adding-repository-custom-instructions-for-github-copilot" class="Link--inTextBlock" target="_blank" rel="noopener noreferrer">Learn how to get started</a>.

[AaronMT]

Might need to bump a point release of the plugin