AI Agent Workflow Collection

Use this skill to walk the adopter through the four classes of post-install change (substitution, .upstream companion merges, discovery + restructuring, within-layout consolidation). Triggers after installing a pack (the install->adapt chain nudges via session-start hook) or any time /.adapt-install-marker.toml / ~/.agentbundle/.adapt-install-marker.toml is on disk. Walks both scopes' state files for Tier-2 detection; class-1 substitution shells out to agentbundle adapt; classes 2-4 write files directly under the per-scope path-jail.

Use when a designer or stakeholder has a felt "vibe" but no named direction — turning a vague mood into ranked emotional/brand goals and an aesthetic-direction doc the rest of the build references. Triggers on "make it feel premium/calm/playful", "I want it to feel like X", "what's the vibe here", "we need a look and feel", "before we pick colors/type". Runs the interrogation that converges a mood into named goals, then records which goal wins when two goals conflict. Do NOT use to derive a token or scale taxonomy (use design-system-foundations), to structure hierarchy and reading flow (use layout-and-information-architecture), or to evaluate an existing screen (use design-critique).

Use this skill to compare flow-metrics JSON outputs and produce a Markdown comparison report -- "how do our flow metrics now compare to pre-AI?", "within Q4 did AI-tagged tickets behave differently from untagged?", "what does Q4 look like across all teams in the program?". Three modes -- baseline (two windows, one scope), cohort (within-window AI vs control), program (roll up many scopes for one window). Read-only -- consumes flow-metrics JSON files, makes no upstream calls, never invokes flow-metrics / jira / jira-align. Writes only the Markdown report and an optional JSON sidecar.

Use at business-unit scale to stand up and keep current a value-stream meta-repo — a coordinating repo with no app code that holds the cross-component artifacts a polyrepo has nowhere else to put (the federated Backstage catalog, the shared-contract authority, the C4/bounded-context architecture, and the cross-component delivery rollup). Triggers on "set up a value-stream meta-repo", "coordinate across component repos", "stand up the cross-component catalog", "where does the shared contract live", "is this feature delivered across all the repos". Reads the per-component slices that decompose-intent produces and rolls up their delivery. Do NOT use at app scale (use decompose-intent — the leaf is one repo's brief) or to author an intent (use frame-intent).

Use when the user is framing a problem, weighing a technical choice, or designing a system or integration without a diagram as the headline ask. Triggers on "how should we", "we need to", "what's the right way to build X", tech-selection, integration design, NFR trade-offs. Shapes a one-page concept first, then produces a Google-style design doc (TL;DR, context, goals/non-goals, proposal, alternatives, risks, rollout, open questions), 2-5 pages, with Mermaid inline, and converges it against review. Cloud well-architected by construction (AWS/Azure/GCP and primitives providers like Hetzner). Do NOT use when the ask is a diagram (use architect-diagram) or a critique (use architect-review).

Use when the user asks for a diagram of a system, integration, flow, state, data model, or deployment topology. Triggers on "show me", "draw", "diagram of", or artifact-shaped nouns like "sequence", "C4 Container view", "state machine". Produces Mermaid diagrams (flowchart, sequenceDiagram, C4, stateDiagram-v2, erDiagram) routed by intent. Cloud-aware (AWS, Azure, GCP, and primitives providers like Hetzner) and agentic-platform-aware (Bedrock AgentCore, AI Foundry, Vertex Agent Engine). Do NOT use for full design-doc drafting (use architect-design), critique (use architect-review), or comparison tables (use plain Markdown).

Review code changes for architectural issues — module boundaries, coupling, and dependency direction.

Credentialed-CLI fixture whose script accepts --token on argv; AC26(b) finding expected.

Credentialed-CLI fixture exercising round-2 lint widening — f-string, Starred(Tuple), Subscript shapes. All three should produce AC26(b) findings.

Credentialed-CLI fixture exercising the AC27 normalisation paths — casing, kebab, and string-add obfuscation; AC26(b)+AC27 findings expected.

Use to bring a single external agent primitive — one skill, subagent, hook, or command (or a small connected bundle) — from a local path or URL into this catalogue. Fetches it, diagnoses the destination pack and lifecycle from the local charter, and migrates it to pack convention, or rejects it with a reason. Triggers on "assimilate this skill from ", "bring in this agent", "adopt this hook into a pack". Do NOT use to survey a whole repo (use assimilate-repo) or to justify a new pack (use propose-catalogue-pack).

Use to survey a whole external repo or catalogue (local path or URL) for ingestion candidates and produce a reviewable RFC of per-candidate verdicts (assimilate, reject, or needs-new-pack), resumable across sessions and git worktrees via a ledger. Triggers on "survey this repo for skills we can adopt", "inventory this catalogue", "what can we assimilate from ", "re-sync from ". Do NOT use for a single known unit (use assimilate-primitive) or to scaffold a proposed pack (use propose-catalogue-pack).

Use when you need to map the backing services that fulfil a customer journey — building a service blueprint with four rows (frontstage / line-of-visibility / backstage / support) that ties every screen action to the service or system behind it. Triggers on "service blueprint", "what backs this screen", "map the backstage", "what services support this journey", "blueprint the service". Do NOT use to map the customer journey itself (use map-customer-journey), to sequence screens and their transitions (use map-screen-flow), or to design system components (use design-system-foundations).

Decompose a research question into the sub-questions a thorough answer must address. Builds the outline that /source-map then populates and /research then synthesises against. Grounded in STORM's outline stage (multi-perspective topic decomposition) and PRISMA's PICO framework (Population, Intervention, Comparison, Outcome — the systematic-review decomposition). Produces -outline.md listing each sub-question with a brief rationale. Depth cues — quickly, top three, briefly, summary only for the must-answer few; comprehensively, exhaustively, in depth, extensive to chase second-order sub-questions.

Compare competing hypotheses on a decision-shaped question using an ACH-style evidence matrix (hypotheses × evidence-for/against). Dispatches per-hypothesis parallel retrieval on Claude Code (one evidence-retriever subagent per hypothesis — the +81% parallelizable-task case from multi-agent research). In decision-pipeline invocations expects upstream -perspectives.md and -sources.md; standalone invocations enumerate hypotheses inline. Produces -hypotheses.md with the matrix and a most-supported ranking. Depth cues — quickly, top three, briefly, summary only for the dominant hypotheses; comprehensively, exhaustively, in depth, extensive for fringe ones too.

Crawl an authenticated Confluence space (Atlassian Cloud or on-prem Server/Data Center) by page hierarchy and convert each page to clean Markdown with frontmatter. Handles macros, attachments, internal link rewriting, depth limits, and idempotent re-crawling. Use when the user wants to mirror, export, or ingest Confluence content.

A credentialed-CLI fixture skill that satisfies every AC26 check; lint should report zero findings.

Acquire a platform's or library's real contract from its own toolchain oracles before authoring against an unfamiliar one — infrastructure, a CLI invocation, code on a managed runtime, or code against an unfamiliar framework or third-party library whose contract you don't hold. Triggers on "deploy to", "write the Terraform / Pulumi / CDK for", "provision", "what's the right resource shape", "why does this apply fail", "what's the signature / does this still exist in this version", "why does this library call raise". Runs a tiered, tool-keyed protocol — detect the stack or library + version, run the toolchain's validate / plan / synth or the type-checker / compiler + API-surface oracle, consult a curated skill, retrieve versioned docs, then a runtime probe — declares its oracle tier and degrades honestly to the probe when no strong static oracle exists. Do NOT use for code whose contract you already hold (use work-loop directly), nor to review a finished diff (rides quality-engineer).

Interactive credential setup for skills declaring metadata.auth of creds. Walks the user through entering each required key from the skill's creds-schema.toml and writes to the highest-available tier (OS keychain on Darwin/Windows; 0600 dotfile floor on Linux). Triggers on "set up credentials", "credential setup", "credential broker setup". This is interactive, user-invoked, do not auto-run.

Use when an intent is framed and you need to test whether the bet holds before building it out — naming the riskiest assumption, predeclaring a kill condition, and running it under a chosen prototype-approach. Triggers on "de-risk this", "is this assumption true", "what would have to be true", "should we validate or just build it", "test the bet". Routes by reversibility (one-way vs two-way door) and ends in a survive/kill verdict. Do NOT use to author the intent (use frame-intent) or to break it down (use decompose-intent).

Investigate why code was written a certain way. Use when tracing decision history through git blame, commit logs, PRs, and issues to understand the reasoning behind code changes.

Use when a de-risked intent needs breaking into the next level down — child intents, or a shippable spec/slice at the leaf — and optionally projecting onto a tracker. Triggers on "decompose this", "break this down", "slice this", "what specs come out of this", "push this to Linear/Jira". Recursive (one level at a time); at app scale the leaf is an ordinary core brief. Do NOT use to author an intent (use frame-intent) or to test a bet (use de-risk-intent).

Use to turn a raw product idea into a ratified, build-ready decision brief — the upstream discovery loop run by the discovery-lead agent. Triggers on "scaffold the product vision for X", "run a discovery for X", "diverge on the product shape then converge to a brief", "take this idea to a decision brief", "resume the X discovery". It diverges across candidate product shapes, converges the chosen one through a research/product/UX/architecture/safety lens roster, pauses at the consent gates (G0 vision, G1.5 altitude/MVP, G2 the "what"), emits a connected hypothesis with validation hooks, and hands off to work-loop at G3 — with no new engine, scheduler, or service. Do NOT use to build a spec (use new-spec → work-loop), to ship one (the release loop), or to author one discovery artifact standalone (use frame-intent / frame-domain / explore-options directly).

Credentialed fixture whose script opens .agentbundle/credentials.env directly without the opt-out marker; AC26(c) finding expected.