AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

named-persona-adversarial-review

Use when a PR has architectural or UX impact, or automated review feels generic, and you want it reviewed through named engineers' documented views (Torvalds, Thompson, Carmack, Beck, Jobs, Cagan). Cites the source for each persona's critique.

Original source
alirezarezvani
License
MIT
Pinned revision
e4a1dc3695908dd5a6c811cc76ab7aa4cf95f2f0

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

Ask your assistant

Use Agentic Memory skills to read alirezarezvani--named-persona-adversarial-review at revision e4a1dc3695908dd5a6c811cc76ab7aa4cf95f2f0, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 2

references/persona_principles.md

Persona Principles — sourced, with confidence levels

This file is the anti-fabrication backbone of the skill. Each persona's review lens must be grounded in a documented principle listed here (or an equally verifiable source you find). Every attribution carries a confidence level:

  • high — the person is documented saying/demonstrating this, with the source named below.
  • moderate — widely attributed and consistent with their body of work, but the exact wording/source is not pinned. Paraphrase; do not put it in quotation marks as if verbatim.
  • low / unknown — you are inferring from their general reputation. Usable as a lens to direct attention, but label it, and never present it as a quote.

Rule: cite the principle, not an invented verbatim quote. If you cannot reach at least moderate confidence for a persona on the point you're making, drop that persona rather than fabricate.


Engineers

Linus Torvalds

  • "Good taste": eliminate the special case. In his 2016 TED talk he walks through a linked-list deletion example where restructuring removes the conditional branch entirely — the special case disappears rather than being handled. Use as a lens on branchy logic and data structures. (confidence: high — TED, "The mind behind Linux," Feb 2016)
  • "Never break userspace." A long-standing, repeatedly stated kernel rule: a change that breaks existing user-space programs is a regression, full stop. Lens on backward compatibility / public APIs. (confidence: high — documented across LKML; e.g. the 2012 "we do not break userspace" thread)

Ken Thompson

  • Trust boundaries — "you can't trust code you didn't totally create yourself." The central thesis of his 1984 Turing Award lecture Reflections on Trusting Trust (compiler backdoor). Lens on supply chain, third-party code, and input you didn't write. (confidence: high — CACM, Aug 1984)
  • Do one thing well. The Unix philosophy of small, composable tools (McIlroy's formulation; Thompson co-created Unix and embodies it). Lens on functions/modules doing too much. (confidence: high for the Unix philosophy; moderate as a direct Thompson quote — it is McIlroy's phrasing)

John Carmack

  • Measure before you optimize; optimization is craft, not guesswork. Consistent across his .plan files, QuakeCon talks, and interviews — profile the real hot path rather than speculate. Lens on performance claims made without data. (confidence: high as a documented stance; moderate on any specific verbatim wording)

Kent Beck

  • Simple design + "make it work, make it right, make it fast" (in that order). From Extreme Programming Explained (1999) and the XP simple-design rules (passes tests, reveals intent, no duplication, fewest elements). Lens on premature complexity and testability. (confidence: high — XP Explained, Addison-Wesley, 1999)

Fred Brooks

  • Essential vs. accidental complexity. From No Silver Bullet (1986) and The Mythical Man-Month (1975): some complexity is inherent to the problem; some is self-inflicted by tooling/design. Lens on whether complexity in the diff is essential or accidental. (confidence: high — IEEE Computer, 1987; MMM, 1975)

Product

Steve Jobs

  • "Start with the customer experience and work backwards to the technology." Stated on stage at WWDC 1997. Also "Design is not just what it looks like — design is how it works" (NYT Magazine, 2003). Lens on error surfaces, onboarding, and complexity the user is forced to absorb. (confidence: high — WWDC 1997; NYT Magazine, Nov 30, 2003)

Marty Cagan

  • "Fall in love with the problem, not the solution." Core theme of Inspired (2008/2017) and Empowered (2020), SVPG. Lens on scope creep, feature-first thinking, and goal misalignment. (confidence: high — SVPG / Inspired)

Des Traynor (Intercom)

  • The first-run experience decides adoption; onboarding and messaging are product. Documented across Intercom's Onboarding writing and his talks. Lens on READMEs, quick-starts, first 30 seconds. (confidence: high as Intercom's documented position; moderate on exact attribution to Traynor personally)

Theory basis (why multiple named lenses work)

  • Edward de Bono, Six Thinking Hats (1985) — deliberately switching perspective surfaces issues a single stance misses. The named personas are "hats" with documented content. (confidence: high)
  • Daniel Kahneman, Thinking, Fast and Slow (2011) — forcing an explicit, effortful (System 2) pass counters the fast, confirmation-biased read. Grounding + the integrity check are System-2 forcing. (confidence: high)
  • Richard Feynman, Cargo Cult Science (Caltech commencement, 1974) — "you must not fool yourself, and you are the easiest person to fool." The integrity check operationalizes this. (confidence: high)

How to use this file during a review

  1. Pick the personas per the Routing table in SKILL.md.
  2. For each, take the principle(s) above as the lens before reading the code.
  3. Every finding names the principle and a confidence level. If your point isn't covered here and you can't verify it by search to at least moderate, either find a persona whose documented principle does cover it, or report the finding on its own technical merit without a name attached.