jonbond5
- 12 skills
- 0 followers
- 8 hours ago last updated
- ▌ Harden Plan · jonbond5Take an implementation plan or design idea that ALREADY EXISTS (usually one Claude just wrote, or a doc/file) and run an adversarial architect-reviewer workflow to attack its assumptions, surface blockers, and iterate it into a hardened plan. Use when you are HOLDING a plan and want it challenged. If you only have a feature idea, use the autobuild plugin's design skill; for a whole project or major refactor, use /project-to-plan.
- ▌ Implement Plan · jonbond5 bundleBuild a plan that already exists. Takes ANY implementation plan (from /harden-plan, /project-to-plan, the autobuild plugin's design skill, or an ad-hoc one) and deploys implementers — in parallel worktrees when the work is disjoint — plus a verification pass that checks every acceptance criterion and then tries to refute each claimed pass. Use when you have a plan and want it BUILT with you in the loop. Does not design — if the plan needs challenging first, run /harden-plan. For unattended build-to-verdict runs, use the autobuild plugin instead.
- ▌ Project To Plan · jonbond5Take a whole-project idea or major refactor and drive it to a hardened implementation plan. Stage A hardens your REQUIREMENTS interactively — reviewer agents challenge the idea and its assumptions, you refine, up to 3 rounds. Stage B turns the hardened requirements into a plan via a non-interactive architect-reviewer loop. Use for a fresh repo, a new system/service, or a major refactor. For a single feature, use the autobuild plugin's design skill; to attack a plan you already hold, use /harden-plan.
- ▌
- ▌ Design · jonbond5Turn a discussion, an idea, or a rough draft into a hardened HANDOFF document that autobuild can build from. Use when the user says "write this up", "turn this into a design", "harden this", "make a handoff", "we're done discussing — capture it", or when a discussion has reached decisions and needs an artifact before anything gets built. Runs an adversarial challenger loop to convergence, verifies load-bearing facts with real commands first, and stops at a document. Does not write feature code.
- ▌ Review · jonbond5Review finished work through multiple independent lenses and return a verdict. Use when code needs reviewing before it lands — a unit an autobuild run just built, a branch, a diff, or a pull request. Runs the lenses in parallel, attacks every claimed pass, and reports findings without fixing them. Not for building, and not for reviewing an idea before it is built.
- ▌ Autobuild · jonbond5Build a feature end to end, autonomously. Use when the user asks to BUILD, IMPLEMENT, ADD, or SHIP something — "build me a feature that does X", "implement X", "add X and make sure it works", "keep going until X is done". Scopes the work into units, builds each in an isolated worktree, attacks every claimed pass adversarially, and stops only when an independent judge says the ORIGINAL ask is satisfied. Expands scope on its own when a discovery blocks the ask. Not for questions, research, or reviewing existing code.
- ▌ Sleep Gate · jonbond5 bundleBridge an agent across a rate-limit window (or other bounded wait) without burning tokens during the wait. Captures durable resume state, runs a pre-sleep checklist, then kicks off `Bash(command="sleep N", run_in_background=true)` so the harness fires a `<task-notification>` on wake. The agent resumes from a breadcrumb file. Use when work continues across a quota reset and the user is stepping away. Do NOT use for context-window pressure, event-driven waits (use Monitor), inside `/loop` (use ScheduleWakeup), or short waits (< ~1 minute).
- ▌ Distill · jonbond5Session-end capture and pruning pass. Scans the current transcript for corrections, stated-rationale decisions, repeated workflows, and self-critique observations; routes load-bearing findings to the right surface (path-scoped rules, always-loaded CLAUDE.md, or auto-memory); reconciles stale status claims in memories and working docs; and enforces a hard budget on every always-loaded surface so it cannot grow without something else shrinking. Runs inline against the active transcript — never delegate to a subagent.
- ▌ Constantly Monitor · jonbond5Set up a recurring, self-healing wake loop that re-invokes you every N minutes to run a user-defined task (poll a log, watch a job, babysit a process), then go idle until the next tick. Uses a recurring CronCreate job as the wake trigger — survives rate-limit windows because each scheduled fire is independent (a wake lost to a quota window is picked up by the next). Use for constant/periodic monitoring where you must keep checking on a fixed cadence. Do NOT use for a single bounded wait (use /sleep-gate), for waking the instant a specific event happens (use Monitor / a backgrounded `until` loop), or for sub-minute cadence (cron is minute-granular).
- ▌ Docker Sweep · jonbond5Find Docker assets this session created that will never be used again — dangling images orphaned by a retag, one-off benchmark images, stopped throwaway containers, BuildKit cache from a one-off build — and report them with the exact narrow reclaim command. Reports by default; deletes only on explicit opt-in. Use at session end after building images or running throwaway containers, or when a partition is filling. Split out of /distill.
- ▌ Close Session · jonbond5 bundleOfficially close the current session — disclose anything still running, write the closing summary, then rename the session to CLOSED_<name> so the agent list shows the thread is finished. Renames both the job state and the session registry, verifies against `claude agents`, and records the close. Reversible with reopen. Use when a session's work is done and you want it filed rather than left open.