chore(ci): narrow the UI consistency check-run agent

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

The Macroscope UI Consistency check-run agent had drifted from a styling guard into a general review and was costing a lot per run. Since #6381 it picked up PR-specific rules from #7906 (contrast tokens, filter on the root) and #7140 (environment routing in shared renderers, which is not a styling concern at all), plus standing instructions to inspect emitted production CSS, trace selector consumers across the codebase, compare cascade layers, and verify runtime scroll behavior on every run.

This cuts the prompt from 11.6 KB to about 4 KB and scopes it back to what it was meant to guard:

1. Use the apps/web/src/components/ui primitives instead of rebuilding controls.
2. Keep styling as Tailwind classes on the owning component instead of adding global CSS or static inline styles.
3. Keep components composable, with the composer banner stack (ComposerBanner.tsx, ComposerBannerStack.tsx, and its consumers) named as the reference for how slot components should be shaped.

The prompt now opens by telling the agent to stay at diff level (no builds, no emitted-CSS inspection, no consumer tracing, no screenshot requests) and that a violation not obvious from the diff is not a finding.

Frontmatter changes: drop git_tools and github_api_read_only, narrow include to .tsx/.css and exclude tests, lower effort from high to medium, and add maxBudgetPerRun: 10. Still runs on claude-opus-5 with conclusion: failure.

The environment-routing rules removed here were a real correctness concern from #7140; if we still want them guarded they belong in their own agent.

Claude Fable 5 via Claude Code

🤖 Generated with Claude Code

---

Note

Low Risk
Config-only change to a CI agent prompt; the tradeoff is less automated coverage for environment routing and deep CSS/theme checks unless another agent picks them up.

Overview
Narrows the Macroscope UI Consistency agent from a broad web review back to a diff-only styling guard for apps/web, with a much shorter prompt (~11.6 KB → ~4 KB) and lower run cost.

The agent now answers three questions only: use components/ui primitives instead of raw controls, keep styling as Tailwind on the owning component (not new global CSS or static inline styles), and follow the ComposerBanner slot pattern for composable UI. It explicitly forbids builds, emitted-CSS inspection, consumer tracing, and screenshot requests; violations must be obvious from the diff.

Frontmatter:effortmedium, maxBudgetPerRun: 10, include limited to **/*.tsx and **/*.css with test files excluded, and git_tools / github_api_read_only removed (still browse_code + modify_pr). showToolCalls dropped.

Removed from the prompt: contrast/token rules, root filter guidance, environment routing in shared renderers, scroll/virtualized-list checks, visual-evidence requirements, and deep CSS cascade/selector auditing—those are no longer part of this check.

^{Reviewed by Cursor Bugbot for commit 04953ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.2 KiB	−93 B (−0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−93 B (−1.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.5 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−58 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−56 B (−0.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 7751299 · PR result: 04953ef · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a CI-only change with no direct product runtime impact, but it materially narrows automated review coverage and modifies a .macroscope/ configuration file. Human review is required for changes that alter Macroscope’s own review behavior.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}