Claude Code and Codex skills are directories, not files
Why Claude Code and Codex discover agent skills as a directory containing SKILL.md, not a flat markdown file, and what silently breaks when an installer gets that wrong.
Ai
Running the blog's content autopilot on a Claude subscription, not API credit
How this blog’s unsupervised publishing pipeline moved off pay-per-token API credit, survived a silent auth revocation, and added a deterministic gate so an unreviewed draft can’t ship as slop.
Two months of senior engineering in 36 AI-assisted hours
A 36-hour window where an autonomous session opened a stack of PRs and the rest of the time went to landing them cleanly.
The pipeline that writes this blog
How this blog’s GitHub Actions workflow mines activity, drafts posts with Claude, and publishes with zero human review.
Bimaaji: agent-safe mutations for Waaseyaa
Why AI agents modifying a Waaseyaa app need a DSL, an AST-safe patch generator, and sovereignty guardrails, instead of raw file edits.
The ingest side of a sovereign language platform
How an Elder’s whiteboard video becomes a searchable Anishinaabemowin lesson, on infrastructure the community owns end to end.
From vibe-coded to shippable: a playbook
Six moves that take a vibe-coded prototype to a state where it can be run, debugged, and handed off.
The hackathon Anthropic didn't expect
A small detail from a Boris Cherny interview points at a bigger story than the one the headlines told.
Agent-friendly JSON output for PHP CI tools
How the waaseyaa/agent-output package shrinks PHPUnit, PHPStan, and bin/check-* output to compact NDJSON envelopes so AI agents do not drown their context window in CI noise.
AI keeps speccing my projects on pre-AI timelines
AI assistants estimate effort using training data from before they existed, then push back when you try to spec faster. Working around it eats hours every week.