test(uipath-troubleshoot): scrub diagnosis hints from legacy scenario manifests

[Stefan-Virgil]

What

Prevents and cleans up diagnosis-hint leaks in agent-visible scenario fixtures.

1. Scrub the two legacy manifests

Removes answer-leaking text from expected_calls[].description in the pre-existing null-reference-exception and argument-null-exception scenarios. The sandbox stages fixtures/, so a thorough agent could read these and skip the diagnosis.

• null-reference-exception: "read the NullReferenceException message and the stack frame pointing at ERN.xaml/CopyFile"
• argument-null-exception: "read the ArgumentNullException stack and InputArguments={}", "confirm the fault at CopyFile", "rule out caller-supplied In arguments (H3)"

Replaced with neutral procedural text (e.g. "Agent fetches the job's execution details."). Only description fields change — rules are untouched, so mock dispatch is identical and no re-validation is needed.

2. Guard against recurrence in the generator

generate_scenario.py now flags the rule in three places so it's caught at generation time, not review:

• module docstring (a dedicated "NO DIAGNOSIS HINTS IN AGENT-VISIBLE FIXTURES" section)
• the emitted manifest _doc (carried into every new scenario)
• the post---apply next-steps output

Scope / safety

• _doc of the legacy manifests left as-is (generic dev notes, not hints). XAML annotations and project.json descriptions were audited and carry no hints. README.md/RESOLUTION.md are not staged to the agent.
• Same scrub already applied to the newer Assign (feat(uipath-troubleshoot): add 6 Assign-originated runtime-exception playbooks + tests #1780) and If (feat(uipath-troubleshoot): add If-condition NRE + KeyNotFound coverage + tests #1782) scenarios.

🤖 Generated with Claude Code

[github-actions]

Claude finished @Stefan-Virgil's task in 1m 37s —— View job

---

PR Review: Scrub diagnosis hints from legacy scenario manifests

• Read review criteria and project conventions
• Read full PR diff
• Read changed files in full context
• Post review

---

Summary

Scrubs answer-leaking text from expected_calls[].description fields in two legacy manifest.json fixtures (null-reference-exception, argument-null-exception). Descriptions that named exception types, fault locations, hypothesis labels, and evidence specifics are replaced with neutral procedural text. This follows the same pattern applied to the Assign and If scenarios in #1780 and #1782.

Change-by-Change Review

1. tests/tasks/uipath-troubleshoot/runtime-exceptions/argument-null-exception/fixtures/mocks/responses/manifest.json

Severity: OK

Six description changes, all correct:

• "read the ArgumentNullException stack and InputArguments={}" → "Agent fetches the job's execution details." — removes exception name and empty-args hint
• "confirm the fault at CopyFile" → "Agent reads error-level job logs for additional execution detail." — removes fault location
• "resolve the Shared folder key" → "Agent enumerates folders to locate the folder that owns the job under investigation." — removes folder name hint
• "rule out caller-supplied In arguments (H3)" → "agent may inspect the process/release configuration." — removes hypothesis label
• Traces description neutralized similarly

All pattern, min, and rules fields untouched — mock dispatch behavior is identical. The _doc field changed from escaped Unicode — to literal —; both are valid JSON representing the same em-dash character (cosmetic, likely from re-serialization).

2. tests/tasks/uipath-troubleshoot/runtime-exceptions/null-reference-exception/fixtures/mocks/responses/manifest.json

Severity: OK

Four description changes, all correct:

• "locate the Shared folder (folder-key 1965a46b-…)" → "locate the folder that owns the job under investigation." — removes folder key hint
• "read the NullReferenceException message and the stack frame pointing at ERN.xaml/CopyFile" → "Agent fetches the job's execution details." — removes exception type, XAML name, and fault method
• "corroborate the stack frame, but the job-get payload already contains the full Info string. min:0 means tracked-but-not-required." → "Agent reads error-level job logs for additional execution detail." — removes stack-frame hint and implementation commentary

Same _doc em-dash encoding normalization as above. All pattern, min, and rules untouched.

What's Missing

Nothing identified. The PR is narrowly scoped to its stated goal and the changes are consistent with the prior scrub PRs (#1780, #1782).

Area Ratings

Area	Status	Notes
Frontmatter	N/A	No skill frontmatter changed
E2E Tests	OK	Fixture-only change; mock dispatch behavior identical
Skill Body	N/A	No skill body changed
References & Assets	OK	Descriptions are neutral and still convey coverage intent
Repo Hygiene	OK	Changes scoped to two files, no secrets, no cross-skill refs

Issues for Manual Review

None found.

Conclusion

Clean PR. The replacement descriptions are neutral and procedural — they preserve the coverage-analysis purpose of expected_calls without leaking diagnosis answers to the agent. The rules and min fields are untouched so mock dispatch is identical. No issues found; this is ready to merge.