The skill — skills/programme/ownership-seam-audit/SKILL.md

name: ownership-seam-audit
description: >-
  Find the decisions in an agentic programme that have no named owner and
  measure what they cost during a simulated incident. Use when org design is
  being discussed, or when an incident stalled waiting for a decision.
allowed-tools: Read, Grep, Glob

Thirty-five of thirty-six hours were two unowned decisions

Org design failures are invisible until an incident, when they appear as hours. The two decisions that typically have no owner — what autonomy rung an agent may run at, and how long traces are retained — are the ones that stall response, because nobody can make them and everybody can veto them.

When to use this

Designing ownership for an agentic programme, and after any incident whose timeline contains a wait.

Procedure

1 — Enumerate the decisions, not the functions. Autonomy rung, trace retention, stop authority, exemption approval, disclosure sign-off, vendor acceptance. Decisions have owners; functions have opinions.

2 — Ask who may decide each one alone. Not who is consulted — who may decide. Two or more names is the same as none, and it takes the same time to resolve.

3 — Simulate an incident that needs several of them. Assign realistic hours, with the unowned decisions taking as long as it takes to convene the people who could make them.

4 — Report the share of the incident that was decision latency. It is usually most of it, and the number is far more persuasive than the observation.

5 — Assign each unowned decision a single accountable name, with an escalation path. Then re-run the simulation to show the difference. That is the business case for the org change.

Example

Input — the fixture committed at the top of scripts/ownership_seam_audit.py. Edit it and re-run: the buckets, counts and verdicts below are derived from it, not hard-coded.

Output — the opening lines of a real run:

seam                        owner                           lesson
------------------------------------------------------------------------------
AppSec ↔ Platform           platform-security               B3.1
Identity ↔ SecOps           on-call SRE, pre-authorised     B3.6
GRC ↔ Engineering           ⚠ NOBODY                        F3.2
SOC ↔ Data                  ⚠ NOBODY                        E1.3
CISO office ↔ Legal         legal, on IR notification       F2.6
AppSec ↔ SOC                detection engineering           E2.2

The run continues past this. test_skills.py executes the script on every build, with no model configured — so what CI proves is that it runs and refuses legibly, not that it prints these lines. The output above is a recorded run against a real model, and nothing re-diffs it: a model's answer is not reproducible, and this repository does not claim otherwise anywhere it can be checked.

Output contract

{
  "decisions": [{"name": "str", "owner": "str|null", "consulted": ["str"], "may_decide_alone": 0}],
  "unowned": ["str"],
  "incident": [{"step": "str", "hours": 0.0, "blocked_on": "str|null"}],
  "totals": {"hours": 0.0, "decision_latency_hours": 0.0, "share": 0.0},
  "after_assignment": {"hours": 0.0}
}

Failure modes

• Mapping functions. They all report coverage.
• Two owners. It resolves as slowly as none.
• Reporting the seam without the hours. It reads as a governance preference.