AGENTS.md@backend/packages/harness/deerflow/agents/task_continuity git:20261010.7e0e301A

Immutable. This exact content never changes and is served at /api/v1/blob/2bd391f690777143.

# Task continuity

When changing `task_note` capacity or batch receipts, follow the
[continuity contract](../../../../../../docs/task-continuity.md) and run the real
graph regressions in `backend/tests/test_task_note_capacity.py`; locking one
snapshot cannot coordinate parallel Commands. With handle resolution enabled,
reservations use the same resolver as execution through the call-local argument
view supplied by `ArtifactResolutionMiddleware`. Both modules import
`RESOLVED_TOOL_CALL_ARGS_KEY` from `state.py`. This view stays in the current
`ToolRuntime.state`, outside messages and checkpoints. Disabled resolution and
direct tool graphs retain raw arguments. Skip non-dict sibling arguments before
reading note fields so malformed calls cannot break valid sibling receipts.
Reservation and execution share `_note_shape_error` for key, content and source
structure checks. Skip structurally invalid calls without source lookups during
reservation; source availability and policy denial remain runtime outcomes whose
unused slots are recalculated in the next batch.

`history_search` accepts optional `role=user|assistant|tool`, mapped at the tool
boundary to stored roles `human|ai|tool`; omission/null keeps all roles. Filter
archived JSON payloads before SQLite FTS `LIMIT 8`, and active messages before
the merged result limit. Preserve ranking, source IDs, returned roles, checkpoint
reachability, and user/thread isolation. Do not add a role parameter to `history_read`.
Sync and async entry points share the implementation; async uses `run_file_io`.
Regression coverage lives in `backend/tests/test_task_continuity.py`.
For usage and trust boundaries, read `docs/task-continuity.md` at the repository root.

Search excerpts are generated by `archive.lookup(excerpts=True)` after result
selection; preserve ranking and the eight-result limit. `excerpt_start` /
`excerpt_end` are zero-based, half-open character offsets in the readable source,
with `source_text[start:end] == excerpt`. Map casefold expansions back to original
characters, not byte offsets. Skip oversized occurrences and continue searching
for a fitting hit, including overlapping active matches, before falling back.
Active sources use substring matching; archived
sources share indexed token spans. If additional FTS normalization prevents exact
location, return the source opening with `excerpt_match=false`. Active text takes
precedence for the same ID. For duplicate archives, use the first rowid version
and its truncation flag, matching reads by ID; offsets must use that same text.
When changing excerpts, read the boundary rules in `docs/task-continuity.md` and
run tests through the actual search/read tools.