feat(context): spill oversized tool output into the sandbox workspace

[devin-ai-integration]

Summary

Completes the per-tool-output story from #880. Today an oversized tool result is truncated to a head+tail preview and the middle is discarded — which can drop the one line that matters (a buried grep hit, a stack frame, a leaked credential). This spills the full output durably into the agent's own sandbox and gives it a bounded preview pointing at the file, so history stays bounded but nothing is lost.

Builds on #880 (per-tool bounding) and #881 (compaction), both merged. Rebased onto main.

Workspace spill only

Every agent run always has a sandbox (SandboxAgent's shell/filesystem require one; the runner always creates it), so the full output is written inside the agent's own sandbox at /workspace/.strix/tool-output/<id>.txt and the agent reads it back with its normal file tools (exec_command / sed / grep / cat) — no dedicated retrieval tool, no host-side store.

async def bound_and_store(text, *, max_lines, max_bytes) -> str:
    parts = _head_tail(text, max_lines, max_bytes)
    if parts is None: return text                       # small: unchanged
    writer = _spill.get("writer")
    if writer is not None:
        path = await writer(uuid4().hex, text)          # -> /workspace/.strix/tool-output/<id>.txt
        if path is not None:
            return _join(head, tail, workspace_notice(path))  # "read it with exec_command ..."
    return _join(head, tail, plain_notice(...))         # write failed: plain preview, bounded, no dangling pointer

Wiring

• strix/core/runner.py: once the sandbox exists, installs a writer via configure_spill_writer(_spill_to_workspace) that does sandbox_session.write(/workspace/.strix/tool-output/<id>.txt, ...), cleared in finally so a later scan can't use a stale session.
• The read_tool_output tool (strix/tools/tool_output/) and the host-side output store (store_full_output / read_stored_output / configure_output_store) are removed entirely.
• _head_tail reserves bytes for the longest possible notice (workspace path or plain) so the joined preview always fits tool_output_max_bytes, and recomputes dropped-line counts after the byte-trim pass.

Test plan

tests/test_output_store.py: small output not spilled; oversized output spilled to the workspace (full text to the writer, notice points at the workspace path, a buried line absent from the preview is present in the spill); multibyte-safe preview within the byte ceiling; writer-failure / no-writer degrade to a plain preview; dropped-line count accounts for byte trimming. Validated live in a real Kali Docker sandbox (215KB output → spill file, agent grep/cat recovers it byte-identical) and in the full integration scan. uv run pytest green; ruff + mypy + bandit pass.

Link to Devin session: https://app.devin.ai/sessions/3461e7dce476402aa2091e4af3f55c00
Requested by: @0xallam

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Greptile Summary

Replaces host-side paginated tool-output retrieval with sandbox-local spill files.

• Wraps oversized string results with asynchronous bounding and persistence.
• Installs a per-scan workspace writer and clears it during runner cleanup.
• Adds coverage for full-output spills, fallback behavior, path notices, and byte bounds.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
strix/tools/output_store.py	Adds bounded previews backed by full-output spill files in the agent sandbox, eliminating the previously reported pagination defects.
strix/core/runner.py	Configures the sandbox-specific spill writer for the scan lifecycle and clears it during cleanup.
strix/agents/factory.py	Converts result bounding to an asynchronous path so every string-valued tool result can be spilled when oversized.
tests/test_output_store.py	Covers spill persistence, bounded notices, multibyte output, and no-writer or failed-writer fallbacks.
tests/test_agent_factory_shell.py	Verifies that assembled function tools retain the result-bounding wrapper.

_{Reviews (15): Last reviewed commit: "refactor(context): workspace spill only;..." | Re-trigger Greptile}

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile

[devin-ai-integration]

@greptile