Skip to content

fix(backend): prevent graph execution stuck + steer SDK away from bash_exec #12548

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
majdyz merged 6 commits into from
Mar 25, 2026
Merged

fix(backend): prevent graph execution stuck + steer SDK away from bash_exec #12548

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
8bb8288
fix(backend/copilot): steer model away from bash_exec for SDK tool-re…
majdyz Mar 25, 2026
d0f558d
fix(backend/copilot): clarify Read tool's dual access in filesystem note
majdyz Mar 25, 2026
3273b95
fix(backend/copilot): revert filesystem note — Read shares sandbox too
majdyz Mar 25, 2026
cbc4bee
fix(backend): guard against missing 'name' in execution input_data
majdyz Mar 25, 2026
9f41880
style: inline 'name' guard into existing condition
majdyz Mar 25, 2026
a6b4cd2
refactor(backend/copilot): deduplicate tool-results guidance to syste…
majdyz Mar 25, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions autogpt_platform/backend/backend/copilot/prompting.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,9 +205,10 @@ def _build_storage_supplement(
### SDK tool-result files
When tool outputs are large, the SDK truncates them and saves the full output to
a local file under `~/.claude/projects/.../tool-results/`. To read these files,
always use `read_file` or `Read` (NOT `read_workspace_file`).
`read_workspace_file` reads from cloud workspace storage, where SDK
tool-results are NOT stored.
always use `Read` (NOT `bash_exec`, NOT `read_workspace_file`).
These files are on the host filesystem — `bash_exec` runs in the sandbox and
CANNOT access them. `read_workspace_file` reads from cloud workspace storage,
where SDK tool-results are NOT stored.
{_SHARED_TOOL_NOTES}{extra_notes}"""


Expand Down
7 changes: 5 additions & 2 deletions autogpt_platform/backend/backend/data/execution.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -342,6 +342,7 @@ def from_db(_graph_exec: AgentGraphExecution):
if (
(block := get_block(exec.block_id))
and block.block_type == BlockType.INPUT
and "name" in exec.input_data
Comment thread
majdyz marked this conversation as resolved.
)
}
),
Expand All @@ -360,8 +361,10 @@ def from_db(_graph_exec: AgentGraphExecution):
outputs: CompletedBlockOutput = defaultdict(list)
for exec in complete_node_executions:
if (
block := get_block(exec.block_id)
) and block.block_type == BlockType.OUTPUT:
(block := get_block(exec.block_id))
and block.block_type == BlockType.OUTPUT
and "name" in exec.input_data
):
Comment thread
majdyz marked this conversation as resolved.
outputs[exec.input_data["name"]].append(exec.input_data.get("value"))

return GraphExecution(
Expand Down
Loading