agentleFS
NeuraCerebra-AINeuraCerebra-AI/Shipworthy/ship-product-workflows/SKILL.md

ship-product-workflows

Skill2 starsChanged 3 months agoMITView on GitHub

Use when safely auditing a product, app, UI, or workflow end to end by trying meaningful user paths and finding user-visible breakpoints across UI behavior, runtime errors, state, persistence, forms, navigation, permissions, API/backend symptoms, responsive behavior, accessibility smoke, and workflow clarity. Trigger for product workflow audits, app QA audits, release-readiness checks, changed-UI regression audits, Playwright or Computer Use click-throughs, 'try every path', 'audit the UI and backend issues', 'where does this break', or 'make sure everything works and looks good' when tied to user-facing workflows. Calls $ship-workflow-clarity for comprehension, consequence, recovery, trust, governance, and human-obviousness lanes. Not for pure code review, pure security scan, backend architecture review, visual polish only, WCAG-only audit, load testing, scraping, or automation without product-workflow audit intent.

The file

Show its rules and text

We split this skill into the rules it gives an agent

17 rules. About 0 are common, 10 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use this skill as the outer product-workflow audit coordinator.

    line 14
  2. Very specific:

    Do not duplicate the full clarity rubric here.

    line 16
  3. Very specific:

    Use a provisional route to load references.

    line 27
  4. Distinctive:

    Do not absorb every reference by default.

    line 31
  5. Distinctive:

    Start with safe-test boundaries and path discovery before judging quality.

    line 44
  6. Distinctive:

    Use code, logs, network traces, database checks, and diffs to explain or verify user-visible symptoms.

    line 47
  7. Very specific:

    Prefer runtime evidence when available.

    line 49
  8. Distinctive:

    Never claim screenshot-only certainty about behavior, accessibility, persistence, state transitions, or unreachable paths.

    line 50
  9. Distinctive:

    Do not click mutating, paid, destructive, permission-changing, privacy-sensitive, publishing, approval, production, or irreversible actions …

    line 52
  10. Very specific:

    Call $ship-workflow-clarity for clarity lanes rather than recreating its full rubric in this skill.

    line 66
  11. Distinctive:

    after input, test Cancel, Back, Close, or navigation after input for loss, warning, and recovery behavior.

    line 80
  12. Distinctive:

    keep one control per distinct input mechanism, distinct invalid → corrected → success boundaries, and separate reload/re-entry proof for …

    line 81
  13. Distinctive:

    treat unexplained disabled controls as findings, unavailable capabilities as feature rows, and false affordances as surface rows.

    line 83
  14. Very specific:

    Use the reference's mechanical identities from the actual route, including surface:<actual-route>:<state>:<role>:<viewport>, and derive …

    line 95
  15. Distinctive:

    Run a provisional router pass: process route, audit mode, evidence path, risk gate, likely tools, and whether a clarity lane is needed.

    line 106
  16. Very specific:

    Call $ship-workflow-clarity for the clarity lane when comprehension, consequence, recovery, proof, governance, or human-obviousness risks …

    line 120
  17. Distinctive:

    Use or adapt templates only when a reusable artifact helps:

    line 128

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Shipworthy Product Workflows

Core Promise

Safely audit product workflows the way a rigorous product QA lead would: discover meaningful user paths, try or trace them with the strongest safe evidence available, find user-visible breakpoints, identify likely fixes, and provide exact verification steps. Treat "every path" as a coverage ambition with honest exclusions, not a promise of omniscience.

Relationship To $ship-workflow-clarity

Use this skill as the outer product-workflow audit coordinator. Call $ship-workflow-clarity as a specialist lane when path evidence should be judged for human-obviousness, comprehension, unclear next actions, consequence/non-consequence boundaries, recovery, trust, governance, accessibility visibility, expert controls, or harmful simplification.

Do not duplicate the full clarity rubric here. Send $ship-workflow-clarity a compact evidence packet after path discovery or runtime tracing, then merge and deduplicate its clarity findings with product/runtime findings.

Router Contract

Route on four axes:

  • Process route: quick, standard, major, or skill_validation.
  • Audit mode: audit_all, audit_selected, audit_top_tasks, audit_high_risk, or audit_changed_only.
  • Evidence path: screenshot/static, repo/diff, browser/runtime, native/desktop, logs/API, design/prototype, mixed, or none-yet.
  • Risk gate: none, low, medium, or high mutation risk.

Use a provisional route to load references. Finalize scope after path discovery. Product validation, such as "check whether this workflow fix works," stays in quick, standard, or major; use skill_validation only when testing this skill's behavior or revisions.

Reference Routing

Do not absorb every reference by default. Load by route and dependency order.

  • quick: one screen, one short path, one screenshot, one small diff, or a known narrow issue. Read references/routing-safety-and-scope.md, then references/reporting-severity-and-validation.md. Add references/runtime-evidence-and-tools.md only when interacting with a live app. Call $ship-workflow-clarity only if comprehension or consequence clarity is materially in scope. Keep output compact: top 1-3 findings, inline coverage/scope, no full template, and no agent/wave plan unless rerouting.
  • standard: one bounded product area, workflow family, or changed feature. Read references/routing-safety-and-scope.md, references/path-discovery-and-coverage.md, references/runtime-evidence-and-tools.md, references/functional-state-and-backend-symptoms.md, then references/reporting-severity-and-validation.md. Add references/living-audit-ledger.md only when the user requests a durable record, file-backed handoff, agents, waves, or compaction protection.
  • major: broad app, multi-role, multi-surface, release-readiness, high-stakes, agent-assisted, wave-based, Playwright-heavy, Computer Use-heavy, Codex macOS, or Claude macOS product workflow audit. Read all standard references plus references/living-audit-ledger.md and references/agent-and-wave-orchestration.md before dispatching agents or tools.
  • skill_validation: testing this skill itself. Read references/reporting-severity-and-validation.md and the route references needed for the forward-test scenario.

Escalate when discovery reveals hidden states, shared components, multiple actors, runtime dependence, unavailable fixtures, medium/high mutation risk, role or permission complexity, AI-to-action handoffs, money/privacy/publishing/destructive actions, or contradictory evidence. Downshift when heavier evidence would not change the answer.

Wrong-route recovery: if evidence shows the route was wrong, state the mismatch, load the missing reference, revise coverage/scope, relabel affected findings, disclose untested paths, and continue. Allow one normal reroute; reroute again only when new evidence materially changes risk, scope, or tooling needs.

Operating Rules

  • Start with safe-test boundaries and path discovery before judging quality.
  • Stay program-agnostic across web, native, mobile, desktop, Electron, game, creative, expert, CLI-like, AI, governed, dashboard, prototype, and internal-tool workflows.
  • Prioritize user-visible truth: what appears, what can be clicked or typed, what happens next, what persists, what errors, and what a user can recover from.
  • Use code, logs, network traces, database checks, and diffs to explain or verify user-visible symptoms. Do not turn this into an unbounded backend architecture audit.
  • For full flagship Shipworthy runs with a runnable UI/app surface, actual frontend path-walking is required. Source, CLI, HTTP, tests, logs, docs, and provider/database probes are supporting evidence, not as a substitute for frontend path-walking.
  • Prefer runtime evidence when available. Static screenshots, docs, and code-only review produce bounded confidence; if no actual frontend path walk occurs, do not call the result a full Shipworthy run.
  • Never claim screenshot-only certainty about behavior, accessibility, persistence, state transitions, or unreachable paths.
  • For fixture-only, screenshot-only, or supplied-diff-only audits, do not blend in ambient workspace/repo context unless the user named it or it is necessary to explain an adjacent risk. If used, label it as supporting context, not as truth about the supplied artifact.
  • Do not click mutating, paid, destructive, permission-changing, privacy-sensitive, publishing, approval, production, or irreversible actions unless the user's authorization covers that action and a verified non-production reset/sandbox contract exists. A disposable fixture alone does not grant authority.
  • For broad, high-stakes, or runtime-heavy audits, use agents, waves, Playwright, browser tools, or Computer Use when available and permitted. If tools or agents are unavailable, report the evidence gap instead of implying they ran.

Browser evidence route

Read this skill's references/runtime-evidence-and-tools.md before choosing a browser evidence path. Use a Codex- or Claude-hosted native browser or computer-use capability as the default for adaptive exploration. Use an existing target-owned Playwright setup only for deterministic replay, explicit assertions, isolated contexts, traces, cross-browser checks, or CI regression proof. Never install Playwright or change the target application merely to obtain browser evidence; record an unavailable capability as evidence debt.

A screenshot proves only the state visible at capture time; it does not prove an entire workflow. Neither native nor Playwright evidence may silently upgrade a finding to Confirmed or verifier status to approved. Include in the audit output the chosen evidence mode, selection reason, observed step boundary, artifacts, limitations, and not-proven statements when this skill runs standalone; return to the orchestrator when lane-dispatched the same bounded evidence context. Shipworthy operates through the four public skills and their skill-owned resources without requiring another product surface.

Browser Failover Gate

Every browser-capable agent instruction must require this behavior: if the native or attached browser locks, disconnects, becomes stale, or cannot create a context, do not stop the lane. Record the failure and switch to an already available independent Playwright process or isolated context; an API such as tab.playwright on the same locked browser binding is not the fallback. If independent Playwright is unavailable or forbidden, return the blocker for another allowed frontend route and leave affected paths unresolved. Never install Playwright, replace runtime proof with static evidence, or claim closure because the browser failed.

  • For major, full-pass, long-running, agent-assisted, or compaction-prone audits, maintain a living audit ledger. Write it to disk when file writes are allowed and useful; otherwise keep it as an explicit report section or proposed artifact path. Treat it as a provisional flight recorder, not a scope lock.
  • Call $ship-workflow-clarity for clarity lanes rather than recreating its full rubric in this skill.
  • Lead with findings. Put scores last, omit them by default, and never use naked scores or decimal grades.
  • Preserve necessary complexity, proof, governance, accessibility paths, expert controls, and productive friction. A "simpler" fix that removes safety or trust is not automatically better.
  • Label assumptions about roles, goals, data, credentials, fixtures, device size, environment, and production risk.

Behavioral Proof Gate

For audit_all and full Shipworthy runs, read references/path-discovery-and-coverage.md and the orchestrator's references/coverage-manifest.md before collection. Freeze independent declared/static/runtime candidate inventories before execution, then build the canonical manifest through exact candidate-to-frontier mappings. Build a material-state control census across every material role, state, viewport, and input mode; enumerate every visible or discoverable interactive control, including duplicate labels, disabled controls, keyboard/context-menu paths, and controls revealed by another action. Record noninteractive false affordances as observed surfaces, never working controls. Copying frontier keys into a census is circular and receives no coverage credit.

Before claiming closure:

  • exercise every safe control once per materially different behavior and record before_state, after_state, and evidence; give unsafe controls a blocked or avoided reason;
  • re-census each spawned surface and each state-changing action across the supplied material variants;
  • inventory discoverable keyboard handlers and probe context-conventional alternatives, including Control/Command+Enter for primary submit, plus command/open, dismiss, activate, and context menu; record positive and negative attempts because omission does not prove absence;
  • after input, test Cancel, Back, Close, or navigation after input for loss, warning, and recovery behavior;
  • keep one control per distinct input mechanism, distinct invalid → corrected → success boundaries, and separate reload/re-entry proof for persistence;
  • split combined observations into distinct findings before verification: one finding, semantic effect code, and exact frontier lineage per independently fixable defect, even when one action or artifact exposes several defects;
  • treat unexplained disabled controls as findings, unavailable capabilities as feature rows, and false affordances as surface rows; and
  • reconcile every raw observation to its control, spawned surface, transition, evidence, and finding before approval. Prose or screenshots cannot replace a missing row.

The apparent-affordance census also includes CTA-like cards, upgrade banners, styled rows, arrow-bearing panels, clickable-looking text, keyboard-only paths, and nonsemantic surfaces that promise interaction. Classify each as functional, informational, unavailable, false affordance, rejected with proof, or out of scope; do not omit it because it is not a semantic control. When requirements promise a path but reconciled source/runtime evidence proves there is no entry point, pending state, or cancellation primitive, record a missing-path finding; genuinely ambiguous behavior remains NOT_PROVEN evidence debt.

Use the reference's mechanical identities from the actual route, including surface:<actual-route>:<state>:<role>:<viewport>, and derive child keys from their parents. Validate the final ledger; invalid shorthand, unresolved observations, or a missing proof row forbids frontier closure.

Execution credit is exact: a receipt must match route, role, state, viewport, containing surface, control identity/type, visibility, enabled state, input mechanism, before state, after state, and evidence linkage. Hidden, disabled, off-route, or instrumentation-generated events can discover a path but cannot prove that it was exercised.

Audit Workflow

  1. Capture the ask, artifact, environment, credentials, fixture data, device targets, safe-test permission, and mutation boundaries.
  2. Run a provisional router pass: process route, audit mode, evidence path, risk gate, likely tools, and whether a clarity lane is needed.
  3. For major, full-pass, long-running, agent-assisted, or compaction-prone audits, start or update a living audit ledger after the provisional route. Record only the safe-test boundary, provisional route, and initial unknowns before discovery; do not pretend the path universe is known yet. Keep newly discovered paths visible instead of treating them as scope creep.
  4. Run the source-backed discovery sequence: freeze declared behavior candidates; freeze static route/component/flag/handler/test candidates; freeze runtime structural candidates; map every raw candidate to the canonical frontier or a reconciliation difference; assign every material row to an owner; then walk the frontend to a spawned-surface fixpoint while re-inventorying after state changes. Build the material-state control census across screens/routes/windows, roles, states, actions, variants, hidden paths, data dependencies, integrations, and mutation risks.
  5. Finalize scope and audit mode:
    • audit_all: audit the discoverable product-workflow surface in scope, with explicit coverage limits.
    • audit_selected: audit user-selected flows, screens, routes, or roles.
    • audit_top_tasks: audit likely frequent, critical, or user-value-driving paths.
    • audit_high_risk: audit money, privacy, security, permission, publish, approval, destructive, AI-to-action, trust, or compliance-heavy paths.
    • audit_changed_only: audit changed routes/components/screens/flows plus adjacent workflows likely affected by the change.
  6. For full Shipworthy runs, return path_frontier additions and frontier burn-down signals after each discovery/testing pass; do not treat newly discovered paths as scope creep.
  7. Decide single-agent, parallel lanes, or staged waves. For major audits, keep the coordinator responsible for safe-test boundaries, evidence standards, deduplication, severity, ledger updates, and final recommendations.
  8. Collect evidence: screenshots, recordings, DOM/UI tree, accessibility tree, focus traversal, console output, network/API observations, logs, route traces, state snapshots, code/diff references, product docs, and path traces. For a full flagship Shipworthy lane, include the actual frontend tool and path-walk status or explicitly downgrade the lane packet.
  9. Execute or trace safe user paths through the actual frontend when available. Exercise every safe control once per materially different behavior. Cover success, empty, loading, error, invalid input, back/forward, cancel, save, refresh, role mismatch, permission denial, responsive, and recovery states when in scope.
  10. Inspect backend/API/data symptoms only where they affect product workflows: failed requests, stale state, missing persistence, authorization leaks, inconsistent payloads, job/status drift, broken imports/exports, or misleading UI after backend failure.
  11. Call $ship-workflow-clarity for the clarity lane when comprehension, consequence, recovery, proof, governance, or human-obviousness risks are present. Pass path evidence and constraints; merge only evidence-backed clarity findings.
  12. Update the candidate inventories, manifest, and ledger after each major discovery pass, material evidence-changing tool batch, agent packet, wave, reroute, major finding, and final synthesis. A lane is not exhausted until every assigned material row is terminal and two independent negative-discovery passes find no new raw candidates.
  13. Deduplicate findings across lanes. Separate confirmed findings, strong findings, provisional hypotheses, non-findings, and untested paths.
  14. Stress-test every fix: what could it break for another role, path, state, device, accessibility route, data condition, or governance boundary?
  15. Report findings first with evidence, severity, confidence, user consequence, likely cause, smallest useful fix, regression risk, and verification step.

Templates

Use or adapt templates only when a reusable artifact helps:

  • templates/audit-report.md: findings-first report skeleton.
  • templates/coverage-map.json: structured coverage inventory.
  • templates/path-trace.md: step-by-step path evidence.
  • templates/finding-ledger.md: issue ledger for deeper audits.
  • templates/evidence-inventory.md: artifact and confidence register.
  • templates/audit-ledger.md: living audit state for major, long-running, or agent-assisted audits.

Output Contract

Lead with findings unless the user asks only for a plan or map. For quick, compress the same concepts into a shorter response.

  1. Top findings or no confirmed findings
  2. Scope, safe-test boundary, process route, audit mode, evidence path, risk gate, tools/agents used or skipped, reroutes, and exclusions
  3. Ledger path, ledger section, or skipped reason when a ledger would normally be expected, plus actual frontend path-walk status for full Shipworthy lanes
  4. Coverage map
  5. Findings by path
  6. $ship-workflow-clarity lane summary when used
  7. Evidence inventory
  8. Assumptions and untested paths
  9. Ranked fix list
  10. Verification plan
  11. Open questions and recommended next pass

For every standalone user-facing final response with unfinished paths, states, controls, proof, exclusions that could be revisited, or blocked/omitted checks, add an exhaustive Remaining Work section directly in the inline chat final response; HTML and artifact links cannot substitute for it. List each stable path/key, status, reason/blocker, next action/proof, and resume condition. End with: “Would you like me to continue through every remaining safe, authorized path and proof item?” followed by “Would you like me to make that continuation a persistent goal so it resumes from the saved checkpoint until the frontier is exhausted?” Put no user-facing prose after those questions. If goal mode is active or unavailable, use the corresponding active-goal or goal-equivalent wording from $ship-readiness-orchestrator. When lane-dispatched, return the complete unresolved register to the orchestrator instead of asking the user.

For each finding, include:

[Severity][Confidence] Path / role-state-device: issue title
Evidence: trace step, screenshot, DOM/UI node, console/network/log entry, code/diff support, or user-test note
Observation: what was seen
User consequence: what breaks, misleads, blocks, risks, or degrades
Likely cause: UI, state, data, permission, API/backend symptom, responsive, accessibility, or clarity
Fix: smallest concrete change
Regression risk: what the fix could break
Verify: exact check to confirm improvement

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.