fix(server): stop dropping attachments on queued follow-ups by rynfar · Pull Request #159 · pylon-code/pylon

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

rynfar · 2026-08-29T05:14:16Z

sendTurn appends an on-disk path line for every attachment. followUp never
did.

Since #140, every adapter except OpenCode deliberately skips non-image
attachments and relies on that line to tell the agent the file exists. So on the
follow-up path a generic file was dropped entirely — not ingested natively,
not surfaced as a path, and nothing errored.

What changes

The path-line construction moves into a shared helper used by both paths. No
behaviour change for sendTurn; followUp gains the lines it was missing.

Not yet reachable, deliberately fixed now

No client can attach a generic file until upstream #8236 (web picker) and
#8237 (mobile picker) land, both still open. That is precisely why this is
worth closing first: the failure is silent, so once those ship it would present
as "the agent ignored my file" with nothing in any log.

Inherited rather than introduced — upstream's ProviderService has the same
shape, with attachmentPathLines only in sendTurn.

Verification

vp test run ProviderService.test.ts PrimeAgentAdapter.test.ts PrimeAgentDaemonAdapter.test.ts — all passing
vp run -F t3 typecheck, vp lint — clean
The routing test adapter had no followUp mock at all, so the service's
follow-up path was unreachable in tests. Added one, then extended the existing
attachment test to assert the path line arrives on that path too.
Mutation probe: reverting to followUpSession(input) fails it with
expected 'and this one' to include '[Attached file "report.pdf" is saved …'

Closes#149.

Claude Opus 5 via Claude Code.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

sendTurn appends an on-disk path line for every attachment; followUp never did. Since every adapter except OpenCode skips non-image attachments and relies on that line, a generic file attached to a follow-up reached the agent as nothing at all - not ingested natively, not surfaced as a path, and no error. Hoists the path-line construction into a shared helper and uses it on both paths. Inherited from upstream, whose ProviderService has the same shape, and not yet reachable in practice because no client can attach a generic file until upstream #8236 and #8237 ship - which is exactly why it should be closed before they do, since the failure is silent. The routing test adapter gains a followUp mock so the service's follow-up path is reachable at all, and the attachment test now asserts the path line arrives there too. Reverting the fix fails it with 'and this one' missing the attachment line. Closes#149.

github-actions · 2026-08-29T05:19:54Z

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	−78 B (−0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+10 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−88 B (−1.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−59 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−59 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	10	9	−1 (−10.0%)	21	✅

Baseline: 71a06e0 · PR result: 329e6fd · 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.

github-actionsBot added vouch:trusted PR author is trusted by repo permissions or the VOUCHED list. size:M labels Aug 29, 2026

rynfar merged commit 2506621 into pylonAug 29, 2026
14 checks passed