| name | sales-daily |
| description | Run the daily sales check. Phase A = invites (every day), Phase B = DM sequencing (Mon-Fri only). Includes pipeline-starvation evaluation. Skipped under --dry-run. |
/sales-daily
Operator-invoked slash command that replaces the daily 9am cron. The
operator runs this each business day; the daily cap policy and
weekend gate are enforced inside the Python entry point.
Step 0 — Verify Attio MCP scope
Before touching any Attio data, confirm read+list+write+delete scope:
- Call
whoami on the Attio MCP — non-error response required.
- Call
list-lists — non-error response required.
- Create a sentinel note on the canary Person record, then delete it.
On any failure, halt and escalate mcp_scope_insufficient directly to
the user. Do not proceed to the daily run if scope is incomplete.
Step 1 — Verify Data Quality halt gate
Before any DM send, check for open P0 alarms in the Operator Review
Queue. Filter type IN (cohort_tagging_regression, write_owner_invariant_violated, migration_idempotency_regression)
AND status='open'. If any row exists, halt with the slug list
visible and tell the operator the recovery path (Attio → Operator
Review Queue → resolve or dismiss the row).
Step 2 — Run the daily check
Invoke:
sales daily
Options:
--dry-run — preview only, skips PB launches AND skips the
pipeline-starvation check (to avoid opening alarm rows during
rehearsals).--force-weekend — override the Mon-Fri DM rule (rarely needed).--skip-dms — invites only.
Weekend policy. Saturday and Sunday: invites are allowed, DMs are
skipped automatically. --force-weekend overrides ONLY the DM gate.
Per the operator-policy memory entry, weekend runs should normally
be invites-only.
The script will:
- Phase 0 — detect newly-accepted connection requests.
- Phase 0.5 — detect responses to DMs.
- Pipeline-starvation check (
evaluate_pipeline_starvation) — opens
a typed queue row if any of the three triggers fire.
- Part A — send connection requests with quarantine + degree-check
gates.
- Part B — DM1/DM2/DM3 sequencing (Mon-Fri only).
LLM dispatch budgets
Per §3.7, industry_classification, borderline_verdict,
synthetic_prescreen, weekly_brain_critique, company_hq_classifier,
synth_holdout, diagnostic_critique, and the Haiku qualifier
execute as Claude Code subagent dispatches from this parent skill —
no Anthropic API key in any engine process. On cost_ceiling_breached,
this skill halts the affected step and logs the typed escalation; the
remainder of the daily run continues if the step is non-critical.
Idempotency
Per-machine flock (workflows/run_lock.py) prevents concurrent runs
of sales-daily on the same machine. The Attio daily_run object
prevents concurrent runs across machines on the same (run_date, machine_id) key. Exit code 75 (EX_TEMPFAIL) on lock contention.
