- Home
- Skills
- Agents & Orchestration
- Engineering Strategy Framework (ESF)
Works with the AI Agents you already use
Engineering Strategy Framework (ESF)
Turns your AI agent into an engineering strategist: it gathers evidence from your code, team and business, tags every claim by where it came from, and returns a ranked plan of risks, debts, quick wins and decisions.
Free
Engineering Strategy Framework (ESF)
Example session with this skill installed
Run an existing-system audit of solidusio/solidus. Use the repository, its git history, the public funding records and the web. Tag every claim with where it came from, rank the risks, and tell me what to do first.
- Read your context and instructions
- Compiled the engineering strategy framework
Audit complete: solidusio/solidus @ cdcdfaf3. This is a condensed version of the report; the full one (28 revisions, a 43-entry decision log) is published at https://vit-panchuk.com/reports/solidus/
In one line: a well-engineered open-source commerce framework that has quietly lost its steward for the second time in ten years — while its main competitor came back with a business model behind it.
Evidence Base — 132 tagged claims
[observed] 51% (69)
[web] 27% (35)
[stakeholder] 17% (22)
[inferred] 5% ( 6)
[assumed] 0% ( 0)
Top risks — ranked by damage × likelihood × whether you'd notice in time
| # | Risk | Would you notice? |
|---|---|---|
| R3 | The half-built new admin becomes a permanent third state: three years in, version 0.4, both admins shipped in every installation | No — every signal a maintainer watches is green |
| R2 | Two firms provide 78% of the money and most of the code | Not for months — a departure looks like a quiet quarter |
| R4 | Spree takes the new-project market | Yes — it is loud and public |
Debts already being paid every release
D1 — Three unfinished rewrites. Promotions, admin and storefront are each designed, written, tested and released twice, with no scheduled end.
D3 — Decisions are never recorded. No architecture document, no decision records, and a roadmap dormant since May 2023.
Easy wins
E1 — Tick the tracker items the code shows are already done, so the issue stops saying the work stopped. Five minutes. [observed]
- E2 — Add Renovate or Dependabot for routine version bumps. An hour.
- E3 — Add bundler-audit to CI, so known-vulnerable versions fail the build. An hour.
Bets for you to decide
| # | Bet | Verdict | Cost |
|---|---|---|---|
| B10 | Publish technical decisions in a short monthly post | Do | an hour a month |
| B6 | Decide the admin's future by gathering the missing evidence first | Decide | one hard screen |
| B7 | Match Spree's React storefront | Kill | — |
Next steps
- Accept, amend or reject the five proposed policies (PL1–PL5).
- Start with E1 — it changes what every evaluator sees today.
- Choose which role Solidus should play, then re-score the bets against it.
Connects securely to your tools. The creator never sees your data.
About this skill
Engineering strategy your agent can actually do — and you can actually check.
Most AI "strategy" output is confident prose you can't verify. ESF works the other way round: the agent collects evidence first, labels every claim with its source — observed in the code, found on the web, told by you, inferred or assumed — and only then recommends. The report shows how much of it is observed and how much is guesswork, so you know what to trust before you act on it.
What you can run it on
- An existing system — a codebase, an engineering team, a business, or even a career.
- A greenfield initiative — before the first commit, when the cheap decisions are still open.
- A single decision — one high-stakes call, worked through with its options and trade-offs.
What you get
- A ranked risk register, a debt ledger, easy wins and strategic bets.
- A decision log and a pre-mortem: how this fails, and what would tell you early.
- Policies the team can adopt, each with who enforces it and when it is reviewed.
- An evidence meter: the share of claims that are observed, sourced, stated, inferred or assumed.
You stay the decision-maker. The agent proposes; you accept, reject or correct, and the corrections are recorded rather than quietly absorbed.
Built on three pillars
- Engineering strategy in the line of Will Larson's Crafting Engineering Strategy.
- Harness engineering: context, guardrails and checks that make agent output trustworthy.
- Continuous improvement (kaizen): two loops — the work itself, and how the work gets better.
Proven on real work. Worked examples are included and published as full public reports: an outside-in audit of the Solidus open-source Rails commerce framework, an anonymised audit of a SaaS ERP for manufacturing, and a greenfield plan. See https://vit-panchuk.com/esf
Free and open source (MIT). Source: https://github.com/vit-panchuk/esf
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires manual approval for policy acceptance. Does not provide numeric simulation or financial modeling (but you can ask your agent to perform it and incorporate it into your ESF report). Best suited for text-based analysis rather than real-time data monitoring.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 21 days ago
- Free with an Agensi account
- Software
- create-spree-app
- Optional
- esf-dsl
Node.js 20+
orBun 1.3+
Git
esf-dsl validates and renders reports and runs via npx. Git is for code audits that read repository history.