The skill — skills/programme/programme-sequencing/SKILL.md

name: programme-sequencing
description: >-
  Simulate a programme's build order against its prerequisites and find the
  steps that block on something not yet done. Use when planning an AI security
  programme, or when a plan starts with evaluation.
allowed-tools: Read, Grep, Glob

Only one step is doable from a standing start

Programme plans are usually ordered by what is interesting or visible. Simulating the order against prerequisites shows which steps block: the popular evaluation-first sequence completes four of six on the first pass and stalls twice, because you cannot evaluate what you have not inventoried or tier what you cannot list.

When to use this

Planning a programme, reviewing somebody else's plan, and when a quarter ended with several things "in progress".

Procedure

1 — Write the steps and their prerequisites. Explicitly, as a dependency list. Most disagreements about order dissolve here, because the dependency is not a matter of opinion.

2 — Identify what is doable from a standing start. Steps with no prerequisites. There is usually exactly one, and it is the inventory.

3 — Simulate the proposed order. Walk it, marking each step done or blocked. Report completions on the first pass and the steps that blocked, with what they were waiting on.

4 — Simulate the dependency-respecting order and compare. Same steps, same effort, different completion — that comparison is the argument, and it does not require anybody to concede a preference.

5 — Say what each order produces at the end of each period. The correct order usually produces nothing demoable for two quarters, and saying so in advance is what stops it being abandoned in the second.

Example

Input — the fixture committed at the top of scripts/programme_sequencing.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:

 step  name          needs     unlocks
------------------------------------------------------------------------------------
    1  inventory     []        you can now tier and assign owners
    2  identity      [1]       per-agent revocation; attribution in logs
    3  containment   [1]       bounded blast radius; a red team has something to test
    4  evidence      [2]       act chains; replayable runs; a drift baseline
    5  evaluation    [4]       accuracy you can defend; regression cases
    6  continuous    [4, 5]    freshness windows; drift alerts; live posture

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

{
  "steps": [{"id": 0, "name": "str", "requires": [0]}],
  "doable_from_start": [0],
  "orders": [{"name": "str", "sequence": [0], "completed_first_pass": 0,
              "blocked": [{"step": 0, "waiting_on": [0]}]}],
  "per_period_output": [{"order": "str", "period": "str", "demoable": false}]
}

Failure modes

• Starting with evaluation. It depends on nearly everything else.
• Arguing about order without the dependency list. It becomes preference.
• Not warning about the quiet quarters. The plan gets abandoned mid-way.