spvce

Use when recording or refreshing what each agent is (mission, owner, runtime, model, tools, skills, permissions) and scoring its persona on the six-point rubric.

Use before any spvce skill reads or writes the shared agent-engineering files (registry, evaluations, improvements, learnings, roadmap), to find or create their home and keep the index.

Use before a proposal is submitted or released, to run a candidate against the baseline and regression suite in isolation and reject regressions.

Use when the owner asks for an audit of their agents, or getting-started runs the first assignment, to produce the five initial outputs read-only.

Use when the owner asks for capability options, an audit names a gap, or the Weekly capability research routine runs, to find top public repos and skills that fill the gap and pass a safety check before anything is proposed.

Use after failure analysis picks a weakness, or when the owner asks for a fix, to draft the smallest targeted change and submit it to the owner through the approval gate.

Use when a bot needs evaluation criteria, test cases, and a baseline (during an audit, before a pilot, or when the owner asks how a bot is doing).

Use when an audit, eval, healthcheck, or the owner needs authorized evidence about a bot (transcripts, routine results, pasted chats, external agent instructions), tagged by source and date.

Use after evidence or a transcript healthcheck shows repeated errors, rework, or waste, or before any change proposal, to cluster them and name one probable cause for the highest-impact proven weakness.

Use on the first message from a new owner (no artifacts index exists yet), or when the owner asks to start over, to run the first read-only audit and set up the agent-engineering program.

Use after a post-release review, a rejected candidate, or a diagnosis worth keeping, to write a reusable lesson and feed it to the roadmap and skill engineering.

Use when evidence, a healthcheck, or owner feedback shows a bot doing a task, to log the task's outcome measures from day one.

Use at the check date in a release's monitoring plan, or when the owner asks how a release went, to compare after vs baseline and recommend keep or roll back.

Use only after "yes, apply <id>" on a proposal with a passing test record, to apply the change with a version entry, monitoring plan, and rollback steps, or to roll a release back.

Use when the audit produces improvement opportunities, or when new results or learnings arrive, to rank the improvement backlog in roadmap.md.

Use when the owner asks for a token-waste audit of every bot's scheduled and event routines.

Use when a diagnosis points to a skill, when the owner asks to improve a skill, or before writing any new skill, to audit a SKILL.md and write a smaller, versioned candidate.

Use during an audit, after a skill release or review, or when the owner asks what skills their bots have, to inventory every skill with version, users, reuse count, and last evaluation.

Use when the owner asks for a friction scan of recent bot chats, to propose skill, bot, or scheduled-job fixes backed by one line of evidence each.

Fires every Tuesday at 10 AM to scan public sources for skills and repos that fill gaps in the owner's agents, reporting only safety-checked, tested finds.