Tooling for AI-assisted software development. Workflows, conventions, and small packages that make working with coding agents (not just talking to them) feel coherent across a project.
| Repo | What it does |
|---|---|
| logmind | Decision logging for AI-assisted development. CLI + GitHub Actions that capture why a change was made, branch-aware, with derived-doc merge drivers so parallel agent work doesn't conflict. Distributed via PyPI + Homebrew. → logmind.dev |
| clud-bug | Claude-powered PR reviewer. One-line install (clud-bug add) gives any repo skill-aware, evidence-anchored code reviews with project-specific rules. → cludbug.dev |
| agent-skills | The shared skill catalog. Skills are markdown files that teach AI agents project conventions — installable via skills.sh, version-pinned per project. The library the other tools draw from. |
| reporulez | GitHub repo defaults: branch rulesets, CODEOWNERS, PR templates, and per-ecosystem Dependabot configs. One command to bring a new repo up to the org's baseline. |
| homebrew-logmind | Homebrew tap for logmind. Auto-bumped on every PyPI release. |
| tokenomics | Decisions + skills repo for the tokenomics product line. |
| rezgen | Application generator / starter scaffolding. |
A new project starts by running logmind init — that wires the decision
log, GitHub Actions, and AGENTS.md guidance into the repo. Then
clud-bug add adds the PR reviewer; on every PR it reads the project's
installed skills (thrillmade/agent-skills) to ground its reviews in
the conventions the team has already declared. reporulez apply
brings the repo up to org baseline (ruleset, CODEOWNERS, Dependabot).
The four together let an AI agent show up in a thrillmade repo and
immediately know how to write decisions, how reviews work, what
conventions to follow, and what gates exist before merge — without any
project-specific setup beyond the four init/add/apply calls above.
- Branch protection: enforced via an org-level ruleset named
org-default-protection(not via per-repo branch-protection rules — that API is legacy and will return 404 for our repos). Seegh api orgs/thrillmade/rulesets. Active settings:- 4 required status checks:
clud-bug-review,check-derived-docs,check-decisions,check-links required_review_thread_resolution: true— unresolved inline review threads block merge- Squash-only + linear-history; non-fast-forward push blocked
- Bypass actor: Repository admin (always) — for self-mod ceremony PRs that legitimately can't satisfy
clud-bug-review
- 4 required status checks:
- Decision log in
docs/decisions.md(anddocs/decisions-branches/<branch>.mdwhile a feature branch is live). - AI-agent guidance in
AGENTS.mdat the repo root.logmind initrefreshes the marker-bracketed block on every install. - Cross-repo notifications: logmind tag pushes notify agent-skills (PR-shape), agent-skills baseline-skill changes notify clud-bug (PR-shape) — both v0.4.0+.
Hobby tier today. Marketing sites: logmind.dev (deployed from logmind/site/), cludbug.dev (deployed from clud-bug/site/). Upgrade trigger: any single repo hitting >20 deploys/day for 3 consecutive days, or onset of the App stream which will likely spike preview deploys. The root vercel.json in both projects uses VERCEL_GIT_PREVIOUS_SHA-aware ignoreCommand to skip deploys when no site/ files changed.