Files
roboco/docs/rag/workflows/escalation.md
T
60c64c70e8 fix(run-hardening): break PM decision-gate, stale-agent, and empty-diff loops (#255)
Forensic triage of a 24h run reconstructed the dominant gateway.rejected
loops from the audit_log. After earlier deploys fixed the i_will_plan crash
and the open_pr push-gap, three real, recurring-capable burn loops remained.
This fixes them at the architecture level, not by prompt-nagging.

journal:decision write-then-gate (the dominant completion-path blocker):
PM decision-point verbs required a separate note(scope='decision') call
before the verb, which loaded/weak models forget to chain — so complete and
unblock hit a tracing_gap (journal:decision missing) and respawn-looped,
stranding finished tasks forever. Each verb now auto-records its OWN
rationale as the journal:decision before the gate runs (the proven
i_am_blocked -> write_struggle pattern), so the gate passes off real,
persisted reasoning. unblock gains a required `reason` (threaded MCP tool ->
request schema -> routes -> choreographer); delegate derives the decision
from its title + description; complete/submit_up/submit_root/escalate_up/
escalate_to_ceo reuse their existing notes/reason. The gate still runs as
defense-in-depth; the auto-record is idempotent within the decision window
and best-effort. Adds JournalService.write_decision and
Choreographer._ensure_pm_decision.

open_pr empty-diff 422: an overlapping-decomposition leaf with zero commits
vs its base makes GitHub 422 "No commits between ...". The generic
invalid_state "retry" looped the dev 15x on one task. open_pr now steers to a
terminal i_am_blocked hand-off so the PM completes or cancels the redundant
leaf.

owns_task stale-agent loop (41x): a superseded agent (task reassigned away)
calling i_am_done/open_pr got a PRECONDITION_OWNERSHIP tracing_gap it read as
a fixable precondition and retried forever. Both verbs now short-circuit with
the clear not_authorized "no longer yours -> give_me_work" steer that
resume/unclaim already use.

RAG docs updated for the new unblock(reason) signature; CHANGELOG entries
added under 0.11.0 (unreleased). open_pr refactored into
_open_pr_preflight_rejection + _open_pr_failure_env to stay within the
return-count and complexity budgets.

Co-authored-by: Renn F <rennf93@users.noreply.github.com>
2026-06-25 01:07:05 +02:00

101 lines
3.0 KiB
Markdown

# Escalation Workflow
## Escalation Chain
```
Developer/QA/Documenter
Cell PM
Main PM
Product Owner / Head of Marketing (Board)
CEO
```
`escalate_up` walks this chain **one rung at a time** — it auto-routes to your immediate escalation target; you cannot choose a higher level or skip a rung.
The one exception is `escalate_to_ceo`: it is a **separate** verb, available only to Main PM and the Board (Product Owner / Head of Marketing), that goes straight to the CEO for final approval of a major task. It is not part of the `escalate_up` chain.
## How to Escalate (up one rung)
```python
escalate_up(
task_id="<task>",
reason="Need clarification on the API contract",
)
```
Auto-routes to your escalation target (you cannot choose it).
## When to Escalate
| Situation | Escalate To |
|-----------|-------------|
| Unclear requirements | Cell PM |
| Blocked by external factor | Cell PM |
| Blocked by another task | Cell PM |
| Cross-cell coordination | Main PM (via Cell PM) |
| Major feature ready for CEO sign-off | CEO (via `escalate_to_ceo`, PM/Board only) |
## Escalate vs Block
| Action | When | Verb |
|--------|------|------|
| **Escalate** | Need a decision / help from above | `escalate_up` |
| **Block** | Can't proceed on an external dependency | `i_am_blocked` |
There is no agent-facing "pause" verb. If you need to step off a task you claimed but haven't progressed, use `unclaim(task_id)` to return it to the pool.
## Blocking a Task
```python
i_am_blocked(
task_id="<task>",
reason="Waiting for the auth service to land",
blocker_type="external",
what_needed="auth-service /token endpoint deployed",
)
```
Your Cell PM is notified and is the one who can `unblock` it.
## CEO Escalation (Main PM / Board Only)
For major tasks requiring CEO approval:
```python
escalate_to_ceo(
task_id="<task>",
reason="Major feature ready for final review",
)
```
Requirements:
- Task must be in `awaiting_pm_review`
- PR must exist
- Only Main PM, Product Owner, or Head of Marketing can call it
- **PARENT TASKS ONLY** — subtasks cannot be escalated to CEO
If you need to escalate a subtask, escalate the parent task instead. The CEO reviews the complete feature, not individual components.
## Good Escalation Format
Include:
- What's the issue
- What context you have
- Specific question
- What you already tried
- How it's affecting work
## Handling Escalations (PM)
1. ACK the notification: `notify_ack(notification_id)`
2. Investigate: read the task, journals, and channel messages
3. Decide, or escalate further with `escalate_up`
4. Communicate the decision (`say` / `dm` / `notify`)
5. Unblock if needed: `unblock(task_id, reason)`
CRITICAL: Verbal resolution is NOT enough. To clear a block you MUST call `unblock(task_id, reason)`. The `reason` (why you are clearing the block) is recorded as your `journal:decision` — no separate `note(scope='decision')` call is required.