← Home / Sectors / Engineering

Sector 09 of 14 — ♐︎ Sagittarius — θ 205.7°–231.4°

Engineering agent trainer

Review, documentation and delivery prompts that are specific to the line and honest about severity. A pack of sixty prompts that trains your AI to work as a staff software engineer carrying out code review, written separately for whichever AI you run.

See a specimen prompt
Prompts
60
Areas of work
6
Tiers
4
AI formats
5

Fig. 01 — The pack map

Six areas of work. Four depths each.

Every area of engineering work is covered at every tier: four Point prompts for quick tasks, three Line, two Plane and one Solid workflow. Six areas of ten makes sixty.

Area of workIPointIILineIIIPlaneIVSolidTotal
01Code review and refactoring432110
02Technical documentation432110
03Incident post-mortems432110
04Release notes and changelogs432110
05Architecture decision records432110
06Test planning432110
Whole pack241812660

Fig. 02 — Specimen

One prompt from this pack, at any depth.

The same task, re-cut live for your AI and for each tier. Watch what each tier adds.

A Department
A Your AI

B Tier

      
Words0
≈ Tokens0
Components0
Variables0

Fig. 03 — The standard

What this pack holds your AI to.

Review is the last check before this code runs in production. A defect missed now costs far more to fix later.

Rules written into the prompts

  • Cite file and line for every finding.
  • Separate must-fix defects from optional suggestions.
  • Do not rewrite the change. Show the smallest fix.

What “done” means

  • Every must-fix names a concrete failure.
  • The verdict is consistent with the must-fix list.
  • A reviewer can check every claim against a named input.

When a person steps in

  • Raise any suspected security vulnerability privately with the security owner before commenting publicly.
  • The AI stops work on that point, says why, and continues with the rest.

Fig. 04 — UK rules

Written for UK firms, under UK law.

Every prompt in this pack tells your AI which UK rules apply to the work, and to apply them only where they are relevant. It never lets the AI claim that work complies with a law: points for a qualified person are marked [REQUIRES VERIFICATION].

UK rules built into the prompts

  • The OWASP Top 10 and the OWASP Application Security Verification Standard
  • The NCSC secure development guidance
  • UK GDPR data protection by design

Specialist viewpoints in every Solid prompt

  • A senior software engineer
  • A security reviewer
  • A site reliability engineer
  • Each judgement is tested from all three, and an explanation the evidence contradicts is removed.

Checks before anything is released

  • Plane and Solid prompts check every input first and list what is missing.
  • Solid prompts stop and ask for a missing input instead of guessing.
  • Solid prompts audit the draft against each UK rule and rewrite any part that fails.

These prompts help your team work to UK rules. They are not legal, tax or financial advice, and a qualified person must review the output before it is relied on.