fix(intake): accept ATIF observation on all step types

[walston]

Problem

Relay emits an observation on any ATIF step, but Intake's schema only permitted observation on agent steps. In-process harnesses like DeepAgents attach an observation to a nested system step, so Intake rejected the entire ATIF POST with HTTP 422 and stored nothing — DeepAgents traces never reached Intake/Studio.

Diagnosed by @yczhang-nv (Yuchen Zhang), who also recommended the fix below.

Fix

• Move observation from AtifStepAgent to the shared AtifStepBase, so every step type (system/user/agent) can carry one.
• Run the observation validators (validate_atif_tool_call_references, validate_atif_v17_subagent_ref_resolution_keys) on all step types (tool-call ids are still sourced only from agent steps).
• Let the mapper read observations on any step (_step_observation no longer gates on AtifStepAgent).
• Add a regression test reproducing a DeepAgents nested-system-step observation.

Verification

• ruff + ty clean; services/intake/tests/test_atif_v17.py 23/23 pass (incl. the new test_atif_v17_observation_on_nested_system_step).
• E2E: with this fix, invoking a DeepAgents agent produces Intake spans including the previously-422ing nested system-* steps — confirmed both via the Intake spans API and in the Studio trace view.

Notes

Independent Intake bug fix, split out from the Fabric email-phishing example work. Getting DeepAgents to emit ATIF additionally requires Fabric ≥ 0207e06 (0.2.0); that dependency bump is a separate, platform-wide change and is not part of this PR.

Summary by CodeRabbit

• New Features

  • ATIF agent, system, and user steps can now include optional observations through the API.
  • Observation data and results are preserved when converted into nested spans.

• Bug Fixes

  • Observation and tool-call references are validated consistently across supported step types.

• Documentation

  • Clarified how execution profiles are filtered and returned across deployment configurations.

• Tests

  • Added regression coverage for system-step observations, references, and mapping behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: a49a251c-2596-495e-a01e-a34d4b1cc6bf

📥 Commits

Reviewing files that changed from the base of the PR and between b6ff20a and 2a38418.

⛔ Files ignored due to path filters (6)

• sdk/python/nemo-platform/.nmpcontext/openapi.yaml is excluded by !sdk/**
• sdk/python/nemo-platform/src/nemo_platform/cli/commands/api/jobs/__init__.py is excluded by !sdk/**
• sdk/python/nemo-platform/src/nemo_platform/resources/jobs/jobs.py is excluded by !sdk/**
• sdk/python/nemo-platform/src/nemo_platform/types/intake/ingest/atif_step_system_param.py is excluded by !sdk/**
• sdk/python/nemo-platform/src/nemo_platform/types/intake/ingest/atif_step_user_param.py is excluded by !sdk/**
• sdk/python/nemo-platform/tests/api_resources/intake/ingest/test_atif.py is excluded by !sdk/**

📒 Files selected for processing (2)

• docs/cli/reference.mdx
• packages/nemo_platform_ext/src/nemo_platform_ext/cli/commands/api/jobs/__init__.py

---

📝 Walkthrough

Walkthrough

ATIF observations are defined on the shared step base, exposed in the OpenAPI schemas, validated for non-agent steps, mapped for all step types, and covered by a nested system-step regression test. Execution-profile documentation now describes capability filtering and topology-specific results.

Changes

ATIF observation handling

Layer / File(s)	Summary
Step observation contracts openapi/ga/individual/platform.openapi.yaml, openapi/ga/openapi.yaml, openapi/openapi.yaml	The schemas now support optional observations on agent, system, and user steps.
Shared observation and validation services/intake/src/nmp/intake/spans/ingest/atif_domain.py	AtifStepBase now stores observations. Tool-call and subagent-reference validation now processes observations on applicable step types.
Observation mapping and regression coverage services/intake/src/nmp/intake/spans/ingest/atif_mapping.py, services/intake/tests/test_atif_v17.py	_step_observation maps observations from all step types. Tests cover nested system-step observations, validation, preservation, and span mapping.

Execution-profile listing documentation

Layer / File(s)	Summary
Execution-profile behavior documentation docs/cli/reference.mdx, packages/nemo_platform_ext/src/nemo_platform_ext/cli/commands/api/jobs/__init__.py	The documentation describes capability filtering, standalone controller pruning, and split-topology API merge results.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 66.67% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: Intake now accepts ATIF observations on all step types.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch intake-atif-observation-all-steps/nwalston

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@services/intake/tests/test_atif_v17.py`:
- Around line 1094-1100: Strengthen the assertions in the trajectory_to_spans
test around the non-agent observation case by locating the nested system-step
span and verifying its mapped output or raw observation attributes contain
“is_likely_phishing: true”. Keep the existing spans presence assertion, but
ensure the test fails when _step_observation returns None.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: ec1b351f-05d2-4e02-926d-00736b7605c6

📥 Commits

Reviewing files that changed from the base of the PR and between e32287b and 44af795.

📒 Files selected for processing (3)

• services/intake/src/nmp/intake/spans/ingest/atif_domain.py
• services/intake/src/nmp/intake/spans/ingest/atif_mapping.py
• services/intake/tests/test_atif_v17.py

[github-actions]

Suite	Lines Covered	Line Rate	Branch Rate
Unit Tests	32004/40616	78.8%	63.5%
Integration Tests	18550/38542	48.1%	20.8%

[coderabbitai]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[walston]

Pushed two follow-up commits to clear the CI blockers (no change to the fix itself):

1. chore(openapi): regenerate spec — moving observation to AtifStepBase changes the ATIF ingest contract, so observation now appears on the system and user step schemas too. lint-openapi regenerates + diffs, so the committed specs were stale. Regenerated via script/generate-openapi-spec.sh (no hand edits): openapi/openapi.yaml, openapi/ga/openapi.yaml, openapi/ga/individual/platform.openapi.yaml — +8/-2 each, all the same observation delta.

2. chore(sdk): regenerate SDK — lint-python-sdk (is-up-to-date) diffs the vendored SDK's .nmpcontext snapshot against the live spec. Ran sdk/stainless.sh sync; result is 5 files: observation added to AtifStepSystemParam/AtifStepUserParam, refreshed .nmpcontext/openapi.yaml, and generated test coverage. is-up-to-date now reports up to date.

One heads-up for review: the SDK regen also swept a docstring-only update in resources/jobs/jobs.py (execution-profiles description) — pre-existing drift on main where the spec text had changed but the SDK hadn't been regenerated. It's generated output, so I left it rather than hand-editing the generated tree. Happy to split it out if you'd prefer this PR stay strictly ATIF-scoped.

[github-actions]

🌿 Preview your docs: https://nvidia-preview-pr-1189-intake-atif-observation-all-steps-nwals.docs.buildwithfern.com/nemo-platform