feat(headless): capture provider stream telemetry

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.

[Astro-Han]

Summary

Capture provider-stream timing, token, interruption, and terminal metadata at the existing host auth proxy; persist one redaction-safe telemetry artifact per Harbor cell; and carry that artifact path through fixed-prompt WAL and infrastructure evidence.

External OpenCode and Kimi Code cells now point traceEventsPath at their actual runtime event stream instead of a nonexistent Maka storage path. This PR observes and preserves evidence only: it does not override Harbor's authoritative task outcome or classify provider failures from telemetry.

Verification

• npm --workspace @maka/headless test — 1,112 passed, 0 failed.
• Focused provider proxy, Harbor runner, and fixed-prompt controller tests — 138 passed, 0 failed.
• Revert fix(headless): expose external agent trace evidence while retaining telemetry — build and 137 focused tests passed.
• Revert feat(headless): capture durable provider stream telemetry while retaining trace evidence — build and 132 focused tests passed.
• git diff --check origin/main...HEAD — passed.

Review focus

The branch contains two independently revertible commits: durable provider telemetry and external-agent trace evidence. It is based on the Harbor lifecycle behavior merged in #1210, including artifact-first recovery after nonzero timeout exits.

The formal Terminal-Bench 2.1 A/B result remains separate in the docs-only report PR.