Each library is one repeated workflow: reusable skills, safe and blocked inputs, review gates, and eval scenarios. Read the public version first, then adapt it before agents get real tools or data.
Each card is a workflow library. Open it for the skills inside, or download everything as one ZIP.
Run the CLI locally, or grab the complete ZIP if you want files without GitHub.
Versioned ZIPs for each library and for the full set, from this site.
Download all Skills (ZIP)A workflow pack for one repeated job: instructions, review checks, safe and blocked inputs, output formats, and eval scenarios. Libraries on this page group those packs by workflow.
Browse workflow librariesPick the workflow closest to your team's repeated job. Each library has multiple skills, review checks, guardrails, outputs, and evals for that workflow.
Technical users can still inspect the full public repository, but nobody has to use GitHub to get the libraries.
One ZIP has every public workflow library. Hand it to your AI tool or whoever owns ops.
Start with the role or workflow category closest to the repeated job your team wants to govern.
Turn account research into a sourced, reviewable brief for sales planning without dumping private notes into AI tools.
Prepare discovery questions, risk checks, buyer context, and follow-up artifacts for complex sales conversations.
Qualify POC requests, protect SE capacity, select the right demo environment, and define success criteria before work starts.
Check campaign intake, consent, segmentation, tracking, personalization, approval evidence, and post-send reporting.
Audit stale playbooks, map approved claims to sources, build role-play scenarios, and capture adoption feedback.
Prepare customer-safe QBR and renewal evidence with health-score freshness, sensitivity stripping, and tone checks.
Turn reply patterns into safer outbound learning loops without leaking private prospect or customer context.
Collect competitive signals, maintain claim evidence, refresh battlecards, QA objection handlers, and route field feedback.
Summarize renewal risk signals, evidence gaps, stakeholder context, and next steps before account reviews.
Move a closed deal into delivery with clean context, risk notes, assumptions, and customer-safe handoff artifacts.
Build shared milestones, owners, blockers, and approval checkpoints for deals or customer implementations.
Triage questionnaire requests, map answers to approved sources, flag risk, and prepare review-ready responses.
Sort security questionnaire items, identify sensitive answer areas, and route claims through approved evidence.
Define input contracts, source labels, acceptance criteria, previews, approval packets, retry budgets, and permission receipts before agents act.
Review context quarantine, permission cards, service identities, capability diffs, tool-result influence, browser profiles, and GitHub inputs before changing agent authority.
Write negative evals, test prompt-injection exposure, replay regressions, review skill lifecycle decisions, and govern self-review writebacks.
Review multimodal evidence, outbound access, trace reuse, assumptions, and boundary changes before agents receive tools or create side effects.
Review task allocation, calibration examples, structured outputs, prompt checkpoints, and resume checkpoints before giving AI workflows autonomy.
Review process evidence, variants, baselines, and intervention choices before assigning AI workflows, Skills, or agents.
Review accountability, tool permissions, disagreement, runnable examples, evals, rollback, and approval before agent launch.
Review rubrics, evidence boundaries, bias probes, disagreements, and authority before an AI judge scores workflow output.
Diagnose stalled AI adoption, pick one workflow, reframe metrics, embed the AI step, and pilot with human review.
Define source, schema, freshness, error, influence, and review boundaries before agents trust tool outputs.
Review proposed durable agent memory for source trust, sensitivity, allowed influence, expiry, rollback, and poisoning risk before future agents retrieve it.
Label AI workflow failures, near misses, and eval misses before teams update Skills, tool contracts, memory, approval gates, or release decisions.
Build a five-case smoke-test packet before reusable Skills, prompts, runbooks, or workflow instructions are shared with a team.
Review whether confidence signals, baseline outcomes, abstention states, and override logs justify more agent authority.
Review no-write agent traces, policy decisions, human corrections, near misses, rollback readiness, and canary scope before agents get write access.
Keep reusable Skills, prompts, and agent runbooks current with owners, dependency drift maps, review triggers, eval paths, and rollback records.
Turn failed AI outputs into mismatch records, failure labels, route decisions, regression cases, and owner-reviewed prompt-change packets before rewriting prompts.
Turn vague agent exception handling into written stop, ask, downgrade, escalate, fail-closed, and resume rules before agents continue with tools.
Turn multi-agent ownership changes into handoff packets with evidence, artifact versions, authority limits, verification status, and receiver acceptance before the next actor continues.
Turn agent help requests into a capacity-aware interruption budget for deciding what interrupts now, what batches, what stays in shadow mode, and what stops.
Turn proposed agent tool access into denied-path tests that prove forbidden tools, targets, arguments, prompt-injected requests, policy failures, and side effects fail before real authority is granted.
Turn temporary agent tool authority into a lease record with expiry, closure, renewal routing, stale replay tests, revocation evidence, and audit gates.
AI Workflows Weekly: guardrails, safer automation, and when new Skills ship.
AI Workflows Weekly
Read archive