WOAFC Toolkit is a Vietnamese-first, evidence-first Codex plugin for auditing products, workflows, interfaces, engineering quality, privacy, market evidence, metrics, and release readiness.
This checkout is the 0.6.0 public-beta release candidate. Public beta begins only after the versioned runtime and cohort gates pass and the immutable v0.6.0 tag resolves to the same commit on both release mirrors. The candidate does not claim production-grade auto-trigger accuracy, compliance certification, or verified support in Claude/Cowork and other hosts.
The research-interview skill in this feature branch is Unreleased and planned for 0.7.0; it is not part of the v0.6.0 candidate or its support claims.
Đọc bằng tiếng Việt · Public beta guide · Skill catalog · Changelog
Use WOAFC when you need:
- An evidence-backed product, workflow, system, or release audit.
- A focused interface, engineering, privacy, cultural-content, metrics, or market-evidence review.
- Realistic user simulation with clear separation from real user research.
- Consent-gated design, live conduct, and evidence-safe synthesis of real qualitative research interviews.
- Context, drift, decision, or remediation diagnosis for an agentic project.
- A saved local report with findings, acceptance criteria, and an implementation plan.
Use the base coding assistant for routine implementation, formatting, simple commands, or fixes with clear acceptance criteria.
After the entry gates pass and the immutable tag is published, choose one release mirror. Both commands pin the marketplace snapshot to v0.6.0; they are not candidate-checkout install commands before that tag exists.
GitHub:
codex plugin marketplace add https://github.com/philngt/afc-toolkit --ref v0.6.0 --sparse .agents/plugins --sparse plugins/woafc-toolkit
codex plugin list --marketplace personal --available --json
codex plugin add woafc-toolkit@personalGitLab:
codex plugin marketplace add https://gitlab.com/earn393021/woafc-toolkit --ref v0.6.0 --sparse .agents/plugins --sparse plugins/woafc-toolkit
codex plugin list --marketplace personal --available --json
codex plugin add woafc-toolkit@personalDo not add both mirrors to the same Codex profile: both intentionally publish the marketplace name personal. If that name already belongs to another marketplace, inspect codex plugin marketplace list --json before changing anything and follow the conflict guidance in the public beta guide.
Start a new Codex thread after installation so the new skill inventory is loaded.
$product-workflow-system-audit audit this repo end to end and give a Go/No-Go recommendation
$product-user-simulation-audit simulate realistic users and find the highest-impact journey friction
$project-steering context-audit this repo before we make the next decision
$research-interview design a discovery interview about churn, then conduct it one question at a time after consent
research-interview supports three bounded modes:
design: create the research brief, participant criteria, consent script, question map, interview guide, and analysis plan.conduct: run a consent-gated interview with a real participant, exactly one question per turn, adaptive branches, shared-understanding checkpoints, and immediateskip,pause, correction, or withdrawal control.synthesize: code eligible transcripts or redacted notes into traceable themes, counter-evidence, limitations, decision implications, and research gaps.
Participation consent and raw-transcript storage consent are separate decisions. The skill never simulates a participant, recommends an answer, or treats interview statements as population or market proof. When the user requests saved artifacts, the study bundle uses .toolkit/research-interviews/YYYYMMDD/<study-slug>/; redacted notes are the default and raw transcripts require explicit raw-storage consent.
ui-usecase-discovery now provides the ui-workflow-audit vertical slice: repository-wide SwiftUI discovery, feature screen graphs, bounded state matrices and journeys, semantic Linux simulation, code-to-UI traces, broken-workflow findings, schema 2.0 reports, and optional native verification packs for an existing UI-test target. It supports iOS and macOS only. Mixed projects skip tvOS, visionOS, and watchOS; projects containing only those targets stop with unsupported_platform. Semantic HTML or browser captures are never presented as native SwiftUI proof.
python3 plugins/woafc-toolkit/scripts/ui_workflow_audit.py run \
--project <swiftui-project> \
--output .toolkit/reports/YYYYMMDD/ui-workflow-audit/| Need | Start with |
|---|---|
| Overall product, workflow, MVP, or release decision | $product-workflow-system-audit |
| Persona journeys and first-use friction | $product-user-simulation-audit |
| Design, conduct, or synthesize real qualitative interviews | $research-interview (Unreleased) |
| Screenshot, Figma, video, running UI, or UI-source quality | $product-interface-audit |
| iOS/macOS SwiftUI graph, state matrix, journeys, simulation, and code traces | $ui-usecase-discovery (Unreleased) |
| Context, drift, root cause, challenge, or durable decision | $project-steering |
| Maintainability, tests, architecture, or data flow | $engineering-quality-audit |
| Privacy, security, retention, permissions, or threat model | $security-privacy-audit |
| Another specialist workflow | See the complete skill catalog |
Explicit $skill invocation is the most predictable release-candidate and public-beta path. Guarded implicit activation exists for narrow, high-confidence audit intent; it is not production auto-trigger proof.
Important audits separate proven, partially proven, user-asserted, assumption, needs-evidence, and blocked. Simulated users are never presented as real market validation. Interview statements remain user-asserted; a transcript proves what a participant said, not that the claim is true or representative.
When a skill is asked to save an HTML report, the default location is:
.toolkit/reports/YYYYMMDD/
Saved HTML reports require a same-folder <report-stem>-implementation-plan.md. Existing reports must not be overwritten without confirmation. .toolkit/ is local and gitignored.
- Public-beta target: Codex CLI
0.144.x; release validation records the exact tested version. - Local release gate: Python
3.10–3.12. - Codex CLI
0.142.5: historical partial runtime evidence for the earlier core flows only. - Other Codex versions, Claude/Cowork, other hosts, and production auto-trigger behavior:
needs-evidenceunless a versioned runtime record says otherwise.
See the compatibility matrix before reporting a host defect.
From the repository root:
python3 plugins/woafc-toolkit/scripts/eval_all.pyThe public-beta gate must end with soft_launch_gate: PASS. A SKIPPED helper is not a PASS; record its reason.
For a published version, the two mirrors are equal release sources only when the same version tag resolves to the same commit. GitHub Issues and GitHub Security Advisories are the authoritative support and security channels; GitLab is a distribution mirror, not a second support queue.
plugins/woafc-toolkit/: installable plugin payload..agents/plugins/marketplace.json: repository marketplace entry.plugins/woafc-toolkit/skills/: skill contracts and references.plugins/woafc-toolkit/evals/: static baselines and scoped runtime evidence.plugins/woafc-toolkit/scripts/: local validation tools.docs/archive/: historical records, not active source of truth.
MIT. See plugins/woafc-toolkit/LICENSE.