[alex-clickhouse]

What

In external mode, each model request of a session that a cron job started carries one X-Nerve-Inference-Cron-Job header. The value names the job. This covers isolated and persistent cron jobs, and the leg session of a workflow run that a cron job started.

How

• nerve/inference_principal.py adds the header next to X-Nerve-Inference-Principal and X-Nerve-Inference-Trigger, with the same rules: local mode sends none, Nerve removes an inherited line from ANTHROPIC_CUSTOM_HEADERS, and a CLI process keeps the value for its whole life. The engine reads the job from the session's origin_detail.cron_job and gives it to the Claude backend in SessionSpec.inference_cron_job.
• Cron job IDs are free text. cron_job_tag_value() changes an ID to 1 to 64 bytes of ASCII letters, digits, ., _ and -: each run of other characters becomes -, and - at the ends is removed. An empty result becomes job-<8 hex of SHA-256>, and a result over 64 bytes becomes its first 55 bytes, - and the hash. So every job gets a value, and two IDs give the same value only rarely.
• The leg session of a workflow run with created_by cron:<job> records the job in origin_detail. Its origin line reads "From cron job ", and its origin link still opens the workflow runs: originLink checks workflow_run before cron_job.
• Nerve's own model calls, "Run later" sessions and Codex send no cron job header. Nerve does not prove the value.

Why

A proxy in front of the model API can attribute model usage by person and by trigger, but not by cron job. Scheduled work is a large part of an agent's model usage, so the cost of each job must be visible.

Testing

• .venv/bin/pytest tests/ (6 parts, temporary HOME): 8686 passed, 39 skipped, total equal to the 8725 collected
• cd web && npm run build && npm test: build clean, 1892 tests passed
• New tests: the transform and its bounds, the CLI environment, isolated, persistent and workflow cron sessions, local mode, and Nerve's own clients (no cron job header)

🤖 Generated with Claude Code

[alex-clickhouse]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Resolved and merged main, preserving both the cron-job and origin headers. Merge commit: 1d92fd3. Focused pytest could not run because pytest is not installed in the checkout.

[nerveai]

Nerve PR review · ⏸ Hold: mostly a duplicate of #658, needs a decision

Since #658 merged, every isolated and persistent cron session already sends X-Nerve-Inference-Origin: cron=<job>. The merge commit 1d92fd3 keeps both mechanisms. So each cron session now names its job twice, in two encodings (cron=Morning_briefing and Morning-briefing), and a proxy has two contracts for one fact. This PR adds two things that #658 lacks. A workflow-run leg that a cron job started names the job; #658 sends only workflow=<run id> for it. A long or non-ASCII job ID gets a hash suffix instead of a plain cut at 56 bytes.

Completeness: tests cover the transform, the CLI env, and cron, persistent and workflow sessions. The docs are in config.md and accounts.md. Claude only, like #658.

Before merge:

• Keep one header for the cron job. Either drop X-Nerve-Inference-Cron-Job (nerve/inference_principal.py:69) and move the leg case and the hash rule into origin_tag, or document why the proxy needs both.
• _leg_origin (nerve/workflows/service.py:457) matches only created_by cron:<job>. A run that a cron session starts with workflow_run_start (session:cron:…, nerve/agent/tools/handlers/workflow_runs.py:202) names no job, but Nerve's per-job usage counts it (nerve/db/usage.py:310). So the proxy's per-job totals would come out lower than the Diagnostics page.

_{Checked: 397 tests passed in an offline sandbox at 1d92fd3 (inference_principal, session_origin, workflow_*, claude_transport). CI had not reported at this head (4/4 ok at ab85e8a). Read the diff, #658's origin_tag and the usage-by-cron-job query. Frontend not run locally · automated review by nerveai — mention @nerveai to ask a follow-up.}

[alex-clickhouse]

Closed: #658 (merged) covers this change. Nerve sends the cron job in X-Nerve-Inference-Origin as cron=<job id>, and the inference proxy tags it as nerve_origin:cron=<job id> (ClickHouse/nerve-private#743). The branch alex/inference-cron-job-header is kept.