jrichlen

Compile deterministic, content-hashed agents from small behavior modules: a typed AgentQuery in, an immutable AgentImage and rendered harness-native subagent out; fails closed on conflicts.

Before writing a new interface (module boundary, class API, function signature, or service contract) that at least two call sites will depend on, that crosses a module/service/team/persistence boundary, or that will be expensive to change later — produce 3+ radically different candidate designs and compare them on depth, locality, and seam placement before picking one. Use for "design it twice".

Map a catalog consolidation by unit of work, not by file: one verb per entity, counts that must reconcile. Use for a "meta diff" or when merging or pruning plugins, skills, services or packages.

Walk the ordered continue, clear, handoff, delegate, compact decision tree at a phase boundary, and keep any handoff artifact pointer-only — settled specs, plans, ADRs, issues, commits, and diffs referenced by path or URL, never copied inline.

Delegate independent work through bounded briefs, owned outputs and checked handbacks. Use when real workers help; keep small or tightly coupled work direct.

Keep a scannable developer journal that writes itself: discover the day's work from sessions, commits and prompts, interview the user, record a dated entry. Ships /dev-diary and /dev-diary-review.

Diagnose a bug by writing ranked, falsifiable hypotheses before any code change, tagging temporary debug instrumentation for a zero-tolerance sweep, and gating the regression test to a red-then-green proof at the confirmed seam. Use when fixing or root-causing a bug.

Audits CLAUDE.md/AGENTS.md/SKILL.md instruction files against current repo state, catches claims that have gone stale (a renamed path, a dropped command, a policy that changed) before they get trusted or acted on, and resolves contradictions between layered instruction files. Use to audit docs or when instruction files contradict each other.

Before any call that transmits repo or user content off-machine (posting a comment, pushing a branch, calling an external API with file contents in the payload), state what is being sent and to whom — permission modes gate the call, this gates the content.

Design and audit an agent system's eval ladder — the cheapest rung that catches each regression, the blind spot beside every green, judges validated by TPR/TNR, pass^k for irreversible actions. Use when designing or auditing a test/eval strategy.

Search for the existing implementation before writing a new one: name the searches run for an existing helper, abstraction, or pattern and what they returned.

Archive old GitHub repositories into a single private graveyard repo as restorable git bundles, then safely delete the originals. Captures full history (branches, tags, PR refs) and never deletes an original until its backup is verified.

Interview the user about a plan before work starts, single-session and no subagents required, walking its design tree and scaling question depth to each branch's stakes (reversibility x blast radius) while offering a recommendation at almost every step. Use on "grill me" or "stress-test this plan".

Scaffold a new marketplace plugin skeleton in one command: valid plugin.json, invariant-first SKILL.md, AGENTS.md symlinks, command stub, and a red-by-default eval, wired into the marketplace.

Close the growth loop's DETECT step: cluster the exhaust every run sheds (stop-reports, scope-fence findings, unmet criteria, diary entries) by failure shape and surface repeat shapes as named candidate invariants. Proposes; never scaffolds. Use when asking what keeps going wrong.

Keep every hunk of the diff traceable to the stated task: out-of-scope discoveries become findings (ticket, note, diary entry), never same-change fixes.

Classify a candidate action as PATCH/MINOR/MAJOR (semver-style blast-radius test) before acting, and rehearse the rollback of any irreversible one — name the restore path and exercise it. Use mid-task when unsure how much autonomy to take.

A halting discipline for iterative fix loops: after a bounded number of failed attempts at the same objective, stop and report state with hypotheses.

Set up and troubleshoot secretless GitHub Actions -> Tailscale auth via Workload Identity Federation (WIF): GitHub OIDC exchanged for short-lived Tailscale tokens, with no stored API keys or OAuth secrets.

Ship the thinnest end-to-end slice through a system first, then widen it in place — for both software delivery and open-ended investigation/research. Use when scoping new work or de-risking unknowns.

Check claims against evidence; optionally pin executable acceptance criteria and obtain independent verification for work that needs a formal contract.

Route every response element to exactly one voice: prose to human-voice, machine-read artifacts to machine-voice; plus ai-writing-mistakes (wording pass) and second-opinion (offer-only validation).

Stress-test a contested thesis with staged, adversarial research (rival hypotheses, ACH matrix, checkpoints), plus fan-out research-and-verify Workflow templates. Not for lookups.

Chart a multi-session effort as a labeled map of typed decision tickets (grilling / prototype / research / task) with explicit dependencies and an open frontier agents self-assign into. Plans; never executes.