Distinguish the native image capability advertisement from the runner's text-only turn contract and assign the missing attachment conversion an explicit P1 follow-up.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Retain primary and cleanup errors together without relaxing durable close checks. Include only bounded process facts and cover cleanup-only failures and credential-safe diagnostics.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Parse ACP MCP namespace and tool names consistently with the TypeScript projection so persisted plan writes anchor to their actual tool boundary. Keep authority and raw argument handling unchanged.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Preserve the independently verified Cursor dependency lock while accepting the prerequisite profile-v3 parser and portable fixture corrections.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Retain forty thousand files and four filename lanes above 32 MiB while fitting the macOS path limit. Allow 300 seconds for real filesystem staging/deletion on macOS and retain the 180-second bound elsewhere.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Record exact image, pack and executable digests with no credentials, network, fixture requests or model prompts. Keep Daytona Product qualification pending.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Record the independently reproduced resolved lock digest after the integrated mainline dependency changes. Candidate image builds continue to supply their own verified lock digest.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Keep the route service factories bound to hoisted test doubles and assert exact calls while preserving both company-boundary 404 checks.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Reject unrelated edits and command mentions, normalize the private workspace path, and replay retained paid evidence offline without another prompt.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Preserve the separate question marker failure and unknown provider USD alongside successful plan approval and controller recovery. Keep the candidate pending.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Preserve the original grader failure and assess the same captured turn through exact tool-call correlation without another prompt.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Reject auto as exact model evidence, release native leases on early workspace failure, and reject pending metadata requests immediately when the provider exits or sends malformed output. Extend early cleanup to the real-service probe.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Handle a missing child PID and broken stdin during early launch failure, retaining the no-prompt failed result and cleanup proof.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Record native permission ID0, denial without filesystem effects, detached async command completion before terminal, process cleanup and externally reconciled unknown credit granularity.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Prepare isolated single-turn denial and detached-command oracles with exact native option identities, marker timing, explicit model admission and process cleanup. Credential-free tests do not claim live qualification.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
