AI Native DevCon 2026 New York — practical agent skills grounded in conference talks
92%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Talk: Version Your Factory, Not Just Your Code
Speaker: Daniel Meppiel
Event: AI Native DevCon New York 2026
Recording: https://www.youtube.com/watch?v=MrjV4uyuBgI
This index paraphrases the supplied corrected SRT. The transcript preserves original cue numbers and start/end timestamps. Cues 193–203 are an inserted conference promotion and are omitted. In Q&A, speaker turns are not reliably labeled; cite the exchange without inventing audience identities.
| Topic | Timestamp | Cues |
|---|---|---|
| Banking-change example and limits of risk classification | 00:44–03:14 | 11–41 |
| Verification, standards, traceability and security | 03:15–04:45 | 42–61 |
| Versioning and digests as identity; software supply-chain analogies | 04:46–06:47 | 62–88 |
| Factories as programs; loose components and portability problem | 06:49–08:53 | 89–115 |
| Contract: assignment, role, dependencies, budget, checks, deliverables | 09:03–10:38 | 118–135 |
| Composition, conditional portability and a factory graph | 10:39–11:36 | 136–148 |
| Demo contract consumes intent/specification and emits diff | 11:49–13:10 | 151–167 |
| APM dependency manifest, lock file, Copilot and OpenCode runs | 13:15–14:52 | 169–191 |
| Promotional interlude omitted | 14:55–15:32 | 193–203 |
| APM description, dependency policy and receipts | 15:39–16:51 | 205–224 |
| Signed receipt, enterprise rules and admission gate | 16:52–18:04 | 225–243 |
| Receipt provenance, check integrity, bill of materials, signer | 18:06–20:22 | 244–276 |
| Gate evaluates evidence, output, policy and signer trust | 20:24–21:04 | 278–287 |
| Accountability and the proposed reduction of review bottlenecks | 21:12–22:02 | 290–299 |
| Trace a production failure; find and revoke compromised skills | 22:04–23:08 | 300–313 |
| Protect checks; version factory when checks change; policy minimum | 23:08–25:13 | 314–343 |
| File-based state, unsigned local audit, signed PR example | 25:13–27:13 | 344–370 |
| Shared format is an open draft proposal | 27:14–28:55 | 371–393 |
| Q&A: deterministic checks, formatting and linting examples | 29:33–30:53 | 405–423 |
| Q&A: trusted evidence avoids rerunning checks centrally | 31:24–31:44 | 426–430 |
| Q&A: shown sequence is one example, not universal waterfall | 31:50–32:54 | 432–448 |
Read the surrounding cues before using these excerpts.
The talk demonstrates a design direction, not a normative schema or an exhaustive security architecture. It does not supply complete command syntax or the linked repositories' URLs in the spoken text. The banking amount in cue 14 and the truncated assignment in cue 153 should not be used as a real payment requirement. The speech alternates between apmx audit and APM audit; do not turn either into verified current CLI instructions.
The signer is described as the running machine/environment in cues 230–231 and 365–367, but as a different machine in cues 262–264. Preserve this uncertainty when discussing implementation. Exact mechanisms preventing an agent from altering checks, controlling signer access, and binding the input environment are not fully specified.
The Q&A around 31:03–31:24 has incomplete phrases and gaps. The later clear explanation concerns avoiding repeated execution of already-attested checks, not removing checks altogether. Individual audience speaker identities cannot be recovered from the subtitles.
.tessl-plugin
skills
talk-boeckeler-context-lifecycles
talk-chanezon-gordon-agent-experience-evaluation
talk-meppiel-versioned-agent-contracts