talk-katsioloudes-code-security-ai

---

name: talk-katsioloudes-code-security-ai description: Use when the user asks about Joseph Katsioloudes's talk "Code Security Reinvented: Navigating the era of AI" — including questions about using AI for security (writing safer code, MCP servers, skills, agentic workflows), the 1-to-100 security-to-developer gap, "start left" vs "shift left", task flows, dual-LLM / LLM-jury, supply chain decisions with AI, AI-assisted fuzzing, hallucinations and non-determinism in AI security review, the GitHub Security Lab's free resources (gh.io/scg, gh.io/sk, gh.io/taskflows), or applying his approach to AI-assisted secure development. metadata: generated-by: talk-to-skill source: file:user-pasted-transcript generated-at: 2026-06-01

Code Security Reinvented: Navigating the era of AI — Joseph Katsioloudes (GitHub Security Lab)

Joseph argues that with only 1 application security specialist per 100 developers, AI is the leverage that can close — or widen — the security gap, depending on whether we use it responsibly. Through a tour of practical demos he shows how to use AI to write safer code, leverage MCP servers and skills, make supply chain decisions, fix vulnerabilities faster in the PR, and educate developers — while being honest about hallucinations, non-determinism, and the limits of AI as a security tool. The throughline: AI is not a replacement for security testing or human-in-the-loop, but it changes the scene, and pairing it with deterministic tooling, good scaffolding, and least-privilege boundaries is what makes it work.

Grounding rules — MUST follow when answering

1. Before answering any specific question, read outline.md to locate the relevant section, then read that section of transcript.md.
2. When attributing words, quote verbatim from transcript.md. Never put quotation marks around paraphrased content.
3. If a claim isn't in transcript.md, say "the talk doesn't address this" — do not infer Joseph's positions from outside knowledge.
4. Cite by transcript line range whenever possible.
5. Speaker attribution is unreliable for this transcript — the source has no per-speaker labels and contains speech-to-text artifacts (e.g. "Macy" the emcee, garbled product names, "Llamas" likely = "LLM-as-judge"). The transcript is almost entirely Joseph speaking, bookended by the emcee's intro/outro and two audience questions during Q&A. Prefer phrasing like "Joseph said..." for the body of the talk, "an audience member asked..." for Q&A, and "the emcee said..." for the framing. Do not invent attributions.
6. Cross-reference any named addressee with the participants list in outline.md before attributing. Where the transcript clearly garbles a term (e.g. "Llamas" → LLM-as-judge, "Copilot Topics" → likely Copilot Autofix), note the likely intended term but quote the transcript verbatim.

How to help with this talk

Apply the speaker's approach to current work

When the user asks "how would Joseph tackle ?" or wants the talk's framework applied to their own situation:

1. Use outline.md → "Named frameworks / concepts" to find the relevant approach (start-left, AI-as-reasoning-layer-on-top-of-deterministic-detection, MCP+skills layering, dual-LLM, task flows, security SLOs, etc.).
2. Read the corresponding range of transcript.md for Joseph's exact wording.
3. Anchor your suggestion in a verbatim quote of how Joseph articulates the framework. Then walk through applying it step-by-step to the user's case.
4. If the framework genuinely doesn't fit the user's situation, say so. Do not stretch Joseph's words to cover cases he doesn't actually address.

Audit the user's situation against the speaker's framework

When the user asks to "audit", "score", "review", or "gap-analyse" their AI-for-security setup against this talk — or describes their situation and asks where they're falling short:

1. Use outline.md → "Named frameworks / concepts" to locate the five areas Joseph covers: (1) writing safer code, (2) MCP servers + skills + agentic workflows, (3) supply chain decisions, (4) remediating alerts faster, (5) developer security education. Also use his MCP-vs-SAST comparison as a sub-framework.
2. For each area, read Joseph's framing in transcript.md and quote it verbatim when stating what "good" looks like.
3. Walk the user through every area in order. If the user hasn't described their state for an area, ask before scoring.
4. For each area, give a clear verdict (covered / partial / missing) grounded in Joseph's criteria, not your own intuition.
5. If an area genuinely doesn't apply, say so explicitly.
6. Summarise gaps at the end and quote what Joseph said about closing them — including the SLO-based education approach he advocates.

Draft an artifact following the speaker's specification

When the user asks to draft an artifact Joseph described — e.g. a supply-chain-decision instruction file, an agents.md, a task flow, an agentic workflow script, security SLOs for a dev team:

1. Locate Joseph's specification in outline.md and the matching range of transcript.md.
2. Capture every constraint Joseph mentions (he is often light on detail — flag this).
3. Before producing the artifact, quote verbatim Joseph's prescription so the user sees the grounding.
4. Point the user to the free open-source templates Joseph names: gh.io/sk (supply chain instruction files), gh.io/scg (hands-on training playground), gh.io/taskflows (vulnerability-finding task flows). Prefer extending those over inventing from scratch.
5. Any parts you add beyond Joseph's prescription, mark clearly as [not from talk — added as a starting placeholder].
6. If the user needs elements Joseph didn't address, say so and ask the user to fill them in.

Factual Q&A about the talk

For any question about what Joseph said, did, or argued:

1. Read outline.md first to find the relevant section(s).
2. Read the matching range of transcript.md.
3. Answer using verbatim quotes from transcript.md. Do not paraphrase Joseph's words while presenting them as a quote.
4. Cite line ranges so the user can verify.
5. If the answer isn't in the transcript, say so explicitly. Do not reach for outside knowledge unless the user asks (and mark such parts as "not from the talk").

Surface this talk proactively when relevant

When the user's current work touches themes Joseph addressed — AI-assisted coding, MCP setup, agentic workflows, security review, supply chain risk, fixing vs detecting — even if they haven't asked about the talk:

1. Briefly note: "Joseph Katsioloudes made a related point in his 'Code Security Reinvented' talk..."
2. Quote verbatim from transcript.md — one quote is usually enough.
3. Add one sentence connecting the quote to the user's situation.
4. Do not over-cite. If the connection is strained, stay quiet.

Teach / explain concepts from the talk

When the user wants to understand a concept Joseph covered (MCP, skills, agentic workflows, task flows, dual-LLM/LLM-jury, fuzzing with AI, start-left, "fixing problem not detection problem"):

1. Look up the term in outline.md → "Terminology glossary".
2. Read Joseph's explanation in transcript.md.
3. Re-explain using Joseph's framing and examples first, with verbatim quotes for the key claims and definitions.
4. You may add modern context, comparisons, or extensions afterwards — but mark them clearly as "not from the talk".

Key quotes

quotes.md contains pre-extracted verbatim highlights from this talk, organised by theme. When formulating answers, check quotes.md first for strong citable evidence before searching the full transcript.md.