My ways of working with coding agents.
setup/init.sh- a plan-then-apply setup script for shared agent docs and labels.setup/github-blockers.sh- audits, syncs, and checks real GitHub blocked-by relationships.setup/templates/*- repo-local issue-tracker, triage, domain, and agent-instruction templates.skills/*- reusable personal skills that can be installed globally by symlink.
Explaining with Video snapshot is an unmaintained reference, not an active skill. To use this workflow, you must install and maintain your own skill separately. Cara's skill installer does not include this snapshot, and Cara does not keep it synced with the maintained source.
Planning and triage are maintainer-initiated steps. /prototype is model-invoked so /wayfinder can use it directly.
/to-speccreates spec issues./to-ticketsbreaks a spec into tracer-bullet tickets with blocking edges./wayfinderplans large work as a shared map of decision tickets on the issue tracker, resolved one at a time./prototypeanswers one unclear logic or UI question with throwaway code, then retains that primary source outside main. Model-invoked so/wayfindercan use it directly./handoffcarries context to another harness, directory, colleague, or side task when continuing in place is not possible./triageevaluates readiness and applies triage labels.
GitHub's issue dependency graph is the canonical blocker source. Markdown
Blocked by: lines are documentation and must not replace real dependency
relationships.
Reference: Skills Changelog v1.2: /wait-what, /writing-for-agents, Claude Code Plugin, and more.
From the target repository:
~/Code/GitHub/zainfathoni/cara/setup/init.shTo apply without the interactive confirmation:
~/Code/GitHub/zainfathoni/cara/setup/init.sh --yesThe script creates or updates repo-local issue-tracker, triage-label, and domain docs, updates the agent-instruction skills block, and creates missing canonical labels. It does not create execution-loop entrypoints.
Markdown Blocked by: #123 lines are documentation only. GitHub's real issue dependency graph is the canonical blocker source. Use the helper to audit or repair repositories where markdown blockers may not have matching GitHub relationships:
~/Code/GitHub/zainfathoni/cara/setup/github-blockers.sh audit --repo OWNER/REPO --state all
~/Code/GitHub/zainfathoni/cara/setup/github-blockers.sh sync --repo OWNER/REPO --state allCheck whether one issue has open real blockers with:
~/Code/GitHub/zainfathoni/cara/setup/github-blockers.sh check-issue --repo OWNER/REPO --issue 168Each configured repository should commit:
docs/agents/issue-tracker.mddocs/agents/triage-labels.mddocs/agents/domain.md
These files adapt global skills to the local repository. They are not an issue tracker or backlog.
After cloning this repo, bootstrap the /sync-skills skill once:
~/Code/GitHub/zainfathoni/cara/skills/install.shThen invoke /sync-skills for all future updates and audits. It fast-forwards this repository, runs both scripts below, cleans up deprecated skills, and verifies the installation.
For manual control or troubleshooting, the underlying scripts are:
~/Code/GitHub/zainfathoni/cara/skills/update-upstream.sh
~/Code/GitHub/zainfathoni/cara/skills/install.shskills/update-upstream.sh installs a release-pinned allowlist of upstream-tracked skills, including shadcn/improve, firewalker06/tycho, and typesafe-ai/skills, and intentionally excludes local-owned teach, so pulling this repo does not depend on a human remembering which upstream skills are safe to update. Upstream wizard remains blocked pending local file/repository mutation hardening.
By default this symlinks the small global selection documented in
skills/README.md into ~/.agents/skills. Pass skill names
to install an explicit project-local or optional selection, and set
AGENT_SKILLS_DIR to install somewhere else.
Upstream skills are installed for Amp, Claude Code, and Codex by default. Set UPSTREAM_SKILLS_AGENTS to a space-separated subset when a machine needs fewer targets.
This repo uses destructive_command_guard scan config in .dcg/hooks.toml and CI in .github/workflows/dcg-scan.yml to catch destructive commands added to setup scripts, skill scripts, and workflows.
For local checks, install dcg through machine configuration when possible (for example, the nix-home Home Manager setup tracked in nix-home#3) and run:
dcg scan --paths setup/ skills/ .github/workflows/ --fail-on error