refactor: consolidate reviewed bot improvements

[jamesadevine]

Summary

Consolidates the strongest correct changes from 45 open bot-authored PRs reviewed
against 559b9188537520d94b442813825699d44f27bdaf.

• Integrates 12 PRs wholly and the unique/correct portions of 10 PRs.
• Deduplicates 18 overlapping PRs.
• Excludes 3 obsolete stale branches and 1 incorrect documentation change.
• Leaves feat(safe-outputs): allow model-selected PR reviewers #2099 open because its head advanced after the frozen review with
  substantial new follow-up changes that are not superseded by this aggregate.
• Leaves all 14 dependency/version bump PRs untouched.

The aggregate includes documentation corrections, audit and safe-output
complexity refactors, CLI/compiler regression coverage, and Conclusion
executor-E2E scenarios.

Validation

• cargo test --all-targets
• cargo clippy --all-targets -- -D warnings
• cargo test --test audit_it --test compiler_tests
• npm test -- --run src/executor-e2e/__tests__/conclusion-cli.test.ts src/executor-e2e/__tests__/runner.test.ts
• npm run typecheck
• npm run build:executor-e2e
• cargo run --quiet -- export-bash-scripts --help confirms --output
• git diff --check
• Azure executor E2E build 641382

The Astro site build fails identically on unchanged main because the locally
cached Starlight version rejects the repository's existing sidebar
autogenerate configuration. The reviewed #2075 site change was independently
rejected and is not included.

Dispositions

PR	Disposition	Result
#2184	Accepted	Canonical container_runtime.rs architecture entry.
#2181	Accepted partially	One-file create_github_issue complexity refactor.
#2179	Duplicate	Superseded by #2184.
#2172	Accepted	set_github_issue_field refactor.
#2167	Accepted partially	add_report_findings helper extraction.
#2163	Accepted	Canonical ADO proxy findings decomposition.
#2161	Accepted	trace --step CLI coverage.
#2160	Duplicate	Superseded by #2184.
#2154	Accepted	Correct permissions-required prompt field list.
#2151	Accepted	ADO script documentation corrections.
#2150	Accepted	Canonical Azure CLI test references.
#2148	Accepted	Correct export-bash-scripts --output documentation.
#2147	Accepted	Correct IR pipeline and Conclusion semantics.
#2145	Accepted partially	Detailed copilot-vendor ecosystem entry.
#2144	Accepted partially	ADO proxy topology and ecosystem-list corrections.
#2136	Accepted partially	Artifact-download helper extraction.
#2125	Accepted partially	Milestone safe-output refactor.
#2123	Duplicate	Superseded by #2184.
#2119	Duplicate	Superseded by #2163.
#2116	Accepted partially	GitHub sub-issue refactor.
#2115	Accepted partially	CompilerExtension::declarations documentation only.
#2114	Accepted	MCP warning-propagation coverage.
#2113	Duplicate	Superseded by #2184.
#2110	Accepted	Conclusion executor-E2E coverage.
#2109	Duplicate	Superseded by #2184.
#2106	Duplicate	Superseded by #2184.
#2104	Duplicate	Superseded by #2184.
#2102	Accepted	Canonical ADO proxy catalog decomposition.
#2101	Duplicate	Superseded by #2184 and #2150.
#2099	Changed after review	Frozen head 4dbcee44 was duplicate of existing functionality; current head 220878c4 adds substantial new integration fixes and remains open for separate review.
#2096	Duplicate	Superseded by #2184.
#2094	Accepted partially	Canonical build-attachment refactor.
#2093	Duplicate	Superseded by #2184 and #2144/#2145.
#2089	Duplicate	Superseded by #2184.
#2081	Duplicate	Superseded by #2102.
#2080	Duplicate	Superseded by #2184.
#2078	Duplicate	Superseded by #2184.
#2075	Rejected	Legacy repositories: example omits checkout and is functionally incorrect.
#2074	Duplicate	Superseded by #2184.
#2025	Duplicate	Superseded by #2102.
#2006	Accepted partially	Isolated create_work_item refactor commit only.
#1946	Obsolete	Stale conflicting create_pull_request rewrite.
#1909	Duplicate	Superseded by #2094.
#1838	Obsolete	Valuable typed-install idea, but unsafe stale implementation.
#1831	Obsolete	Valuable package-feed idea, but unsafe old-architecture branch.

Review safeguards

• Every disposition was made against a frozen head SHA.
• All 45 head SHAs and origin/main were rechecked before commit; none changed.
• The closure-time SHA guard caught feat(safe-outputs): allow model-selected PR reviewers #2099 advancing after review; it was not
  closed or represented as superseded.
• Duplicate families use one canonical implementation.
• Partial integrations omit unrelated cumulative commits and duplicate edits.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
1 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[github-actions]

Prompt evaluation

Note

This is an advisory static review. Only Prompt Contracts is merge-blocking.

Only prompts/create-ado-agentic-workflow.md changed; the update and debug
prompts and the shared contract are byte-identical between base and head, so
only the create suite is in scope. The only textual change is adding
permissions-required to the "optional" front-matter key list in step 2
(Build Front Matter). The field is real (src/compile/types.rs,
docs/front-matter.md) but per the doc comment is "usually set by an imported
component rather than authored directly," and none of the three create cases
involve imports, so it has no material effect on any case's expected output.

Prompt	Cases	Improved	Unchanged	Regressed	Inconclusive
create	3	0	3	0	0

Potential regressions

None identified.

Per-case scores

Case	Prompt	Base	Candidate	Result
create-minimal-manual	create	16/16	16/16	unchanged
create-scheduled-workitem-report	create	16/16	16/16	unchanged
create-needs-clarification	create	16/16	16/16	unchanged

Scores are the sum of 4 common criteria (task_completion, grounding,
safety_and_consent, clarity_and_done_criteria) + 4 create criteria
(create_workflow_coherence, create_trigger_scope,
create_tools_outputs_permissions, create_no_action), each 0-2, against the
prompt text (this is a static instruction review, not an executed run).

• create-minimal-manual: Both prompt versions instruct a manual trigger,
  minimal fields, no safe-outputs, and an explicit "No Action" section mapped
  to summarizing zero matches — matches ground truth (trigger: manual,
  safe_outputs_required: false). The added permissions-required key is
  irrelevant here (no imports).
• create-scheduled-workitem-report: Both versions require a compact front
  matter with tools, permissions, and safe-outputs, and an explicit
  "No Action" section mapped to noop — matches ground truth (schedule,
  comment-on-work-item max 3, target). permissions-required is not implicated
  since the reference config uses concrete permissions.read/write, not
  imports.
• create-needs-clarification: Both versions' step 1 ("Gather Required
  Inputs" / "ask only missing essentials first") and Done Criteria ("When
  essential inputs are missing, concise clarification is returned instead of
  an invented workflow") are unchanged and drive the expected clarification
  outcome identically.

Generated by Prompt Evaluator for #2189 · auto · 28.8 AIC · ⌖ 1.6 AIC · ⊞ 8.3K · ◷