Til

Learned

A personal knowledge base following the [Zettelkasten](https://en.wikipedia.org/wiki/Zettelkasten) methodology, maintained by LLMs using the [LLM Wiki](https://gist.github.com/karpathy/1dd0294ef9567971c1e4348a90d69285) pattern.

Interlinked notes organized by topic. Browse in [Obsidian](https://obsidian.md/) for graph view and backlinks, or read the [full index](index.md).

Recent additions

• 2026-07-18 macOS LaunchAgent/LaunchDaemon authoring for a BTM-friendly identity - ProgramArguments[0] must be the launcher's own path, no .sh extension on the entry point, #!/bin/bash not env bash so TCC grants survive a bash upgrade
• 2026-07-18 1Password backup pattern for Apple developer signing certs - Split a codesigning .p12 + its passphrase into two tagged 1Password items (Document + Password) so a lost keychain doesn't lose the cert; restore flow and the security export -t identities scoping caveat
• 2026-07-04 Scaling the harness: six components of an agentic system - P = f(R,M,C,S,O,G); three named failure modes (exposure-without-access, stale-but-confident, confident-but-unchecked) and the process-metrics evaluation agenda
• 2026-07-04 Finding your unknowns: the bottleneck in agentic coding - Quality is bottlenecked by how fast the operator clarifies unknowns; a 2x2 with a distinct technique per quadrant (blindspot pass, interview, prototype-and-react)
• 2026-06-26 Secret resolution for pi agent providers via 1Password op read - !op read and $ENV_VAR keep provider API keys out of plaintext auth.json/models.json; service-account auth makes the resolution headless and transcript-safe
• 2026-05-26 [LKY operati