FDE AI Systems Reference Implementations

Three runnable enterprise AI systems demonstrating secure RAG, governed agents, AI release reliability, evals, traces, audit logs, and approval gates.

Most AI app demos stop at chat. Real enterprise deployments need permission boundaries, evidence, human approval, release reliability, debugging surfaces, and regression tests. This repo implements those patterns in three local-first systems that run without paid APIs, while leaving clean upgrade paths to OpenAI Responses API, Agents SDK, PostgreSQL/pgvector, OpenTelemetry, and enterprise connectors.

Projects

Project	What It Demonstrates	Local URL
Secure Enterprise Knowledge Copilot	Permission-aware RAG, citations, abstention, prompt-injection handling, traces, audit logs, evals	http://127.0.0.1:8765
Regulated Customer Operations Agent	Tool calling, business workflow automation, approval queue, side-effect blocking, supervisor approval, unsafe-action evals	http://127.0.0.1:8770
AI Reliability Incident Console	Canary release triage, eval regression evidence, rollout blocking, remediation plans, traces, audit logs	http://127.0.0.1:8780

Architecture cards:

Read these with Architecture Boundaries, API Contracts, and Runtime UI Contracts.

Project	Backend Boundary	Browser Boundary	Core Safety Controls	Fast Evidence
Secure Enterprise Knowledge Copilot	secure-enterprise-knowledge-copilot/src/copilot owns retrieval, answering, storage, security, evals, and the opt-in model gateway.	secure-enterprise-knowledge-copilot/web/js keeps local ES modules for API calls, rendering, trace links, theme, clipboard, and scenario drafts.	Permission filtering before generation, citation-required answers, abstention, prompt-injection handling, traces, audit logs.	Run python -B scripts/dev.py smoke and inspect the Alice/Morgan finance path in the Demo Path Map.
Regulated Customer Operations Agent	regulated-customer-operations-agent/src/ops_agent owns workflow routing, governed tools, storage, evals, and the opt-in model gateway.	regulated-customer-operations-agent/web/js keeps local ES modules for case investigation, approvals, trace links, theme, clipboard, and scenario drafts.	Side-effect tools require application approval, bypass attempts are refused, supervisor execution is auditable.	Run python -B scripts/dev.py smoke and inspect the Ivy case-1001 approval path in the Demo Path Map.
AI Reliability Incident Console	ai-reliability-incident-console/src/reliability_console owns release state, incident triage, eval evidence, storage, and audit records.	ai-reliability-incident-console/web/js keeps local ES modules for incident selection, triage rendering, trace links, theme, clipboard, and scenario drafts.	Failed eval evidence blocks unsafe rollout, remediation stays traceable, audit events link release decisions to incidents.	Run python -B scripts/dev.py smoke and inspect the unsafe canary release path in the Demo Path Map.

Risk badges:

Project	Control Badges	Primary Evidence
Secure Enterprise Knowledge Copilot	permissions citations abstention prompt-injection handling evals traces audit logs	Evidence Matrix, Threat Model, Observability Integrity
Regulated Customer Operations Agent	tool governance approvals side-effect blocking supervisor review evals traces audit logs	Evidence Matrix, Threat Model, Observability Integrity
AI Reliability Incident Console	eval-regression evidence release blocking remediation planning incident triage traces audit logs	Evidence Matrix, Threat Model, Observability Integrity

Why This Exists

FDE and AI application systems need more than a model call. These reference implementations focus on the controls that usually separate production-oriented AI systems from chatbot demos:

• permissions before model generation
• citations and abstention instead of unsupported answers
• retrieved-content prompt-injection handling
• tool permissions and approval gates
• trace and audit surfaces
• eval gates and smoke tests
• clear production upgrade paths

Quickstart

Verify everything from a clean checkout:

python -B scripts/dev.py verify

Start all demos:

python -B scripts/dev.py start

Or start them separately:

cd secure-enterprise-knowledge-copilot
python -B app.py --reset --port 8765

cd regulated-customer-operations-agent
python -B app.py --reset --port 8770

cd ai-reliability-incident-console
python -B app.py --reset --port 8780

Command quick-reference:

Workflow	Commands
Local run	python -B scripts/dev.py start
Verification	python -B scripts/dev.py verify, python -B scripts/dev.py quality, python -B scripts/dev.py smoke, python -B scripts/dev.py demo-presets, python -B scripts/dev.py evals, python -B scripts/dev.py contracts, python -B scripts/dev.py safety
Release evidence	python -B scripts/dev.py replay-artifact, python -B scripts/dev.py report, python -B scripts/dev.py readiness-report, python -B scripts/dev.py fresh-clone, python -B scripts/post_publish_check.py
Visual assets	python -B scripts/dev.py visual-assets, python -B scripts/dev.py visual-asset-diff, python -B scripts/dev.py refresh-visual-assets
GitHub maintenance	python -B scripts/dev.py github-readiness, python -B scripts/dev.py pr-triage, python -B scripts/dev.py github-maintenance, python -B scripts/dev.py github-community
Optional environment checks	python -B scripts/dev.py container-release, python -B scripts/dev.py docker-runtime, python -B scripts/dev.py openai-live

Command decision tree:

Need	Start With	Follow With
Prove the local repo works from a normal checkout.	python -B scripts/dev.py verify	Use the command output expectations table below if a gate fails.
Review a code or docs change before publishing it.	python -B scripts/dev.py quality	Run python -B scripts/dev.py fresh-clone-local when the change affects public setup, docs, assets, or runtime paths.
Prepare release-facing evidence after a push.	python -B scripts/dev.py fresh-clone	Run python -B scripts/post_publish_check.py, then python -B scripts/dev.py github-readiness.
Check screenshots or frontend visual drift.	python -B scripts/dev.py visual-assets	Use python -B scripts/dev.py visual-asset-diff; refresh with python -B scripts/dev.py refresh-visual-assets only when screenshots intentionally change.
Review GitHub state or public PRs.	python -B scripts/dev.py pr-triage	Use python -B scripts/dev.py github-maintenance and python -B scripts/dev.py github-community for dry-run account setup or community sync plans.
Check optional environments.	python -B scripts/dev.py container-release	Run python -B scripts/dev.py docker-runtime only on Docker-enabled machines, and python -B scripts/dev.py openai-live only in an API-key environment.

Full command index:

python -B scripts/dev.py assets
python -B scripts/dev.py api-docs
python -B scripts/dev.py architecture
python -B scripts/dev.py claims
python -B scripts/dev.py community-issues
python -B scripts/dev.py container-release
python -B scripts/dev.py docker-runtime
python -B scripts/dev.py dependency-surface
python -B scripts/dev.py demo-presets
python -B scripts/dev.py contracts
python -B scripts/dev.py error-hygiene
python -B scripts/dev.py frontend
python -B scripts/dev.py health
python -B scripts/dev.py evals
python -B scripts/dev.py eval-csv
python -B scripts/dev.py github-launch-setup
python -B scripts/dev.py github-community
python -B scripts/dev.py github-maintenance
python -B scripts/dev.py fresh-clone-local
python -B scripts/dev.py fresh-clone
python -B scripts/dev.py github-readiness
python -B scripts/dev.py governance
python -B scripts/dev.py launch-assets
python -B scripts/dev.py model-gateway-safety
python -B scripts/dev.py observability
python -B scripts/dev.py openai-live
python -B scripts/dev.py otel-traces
python -B scripts/dev.py pr-policy
python -B scripts/dev.py pr-triage
python -B scripts/dev.py readiness-report
python -B scripts/dev.py refresh-visual-assets
python -B scripts/dev.py replay
python -B scripts/dev.py replay-artifact
python -B scripts/dev.py scenario-data
python -B scripts/dev.py smoke
python -B scripts/dev.py report
python -B scripts/dev.py safety
python -B scripts/dev.py threat-model
python -B scripts/dev.py ui-contracts
python -B scripts/dev.py visual-assets
python -B scripts/dev.py visual-asset-diff
python -B scripts/dev.py workflow-security
python -B scripts/dev.py quality
python -B scripts/post_publish_check.py

Command output expectations:

Command	Success Signal	Notes
python -B scripts/dev.py verify	Starts or reuses the three local services, runs the CI-quality gate, and ends with Quality gate passed.	Use before local release review when the demo services should be exercised.
python -B scripts/dev.py quality	Runs repository safety, docs/assets, UI contracts, service health, smoke flows, evals, replay artifacts, and claim checks; ends with Quality gate passed.	This is the main local quality gate.
python -B scripts/dev.py demo-presets	Validates docs/demo_state_presets.json against seed and eval data; ends with Demo state presets check passed.	Use before recording, reviewing, or sharing canonical local demo paths.
python -B scripts/dev.py fresh-clone-local	Clones the current checkout into out/fresh-clone-tmp/, runs release-facing checks, starts isolated demo ports, and ends with Fresh clone experience check passed.	Use before push when the remote branch may not include the current commit yet.
python -B scripts/dev.py fresh-clone	Clones origin, runs the same fresh-clone checks, starts isolated demo ports, and ends with Fresh clone experience check passed.	Requires network access and a pushed commit.
python -B scripts/post_publish_check.py	Prints [PASS] rows for the GitHub page, raw README/workflow, and required published files; ends with Post-publish check passed.	Use after push to confirm public GitHub assets are reachable.
python -B scripts/dev.py github-readiness	Prints [PASS], [WARN], or [MANUAL] rows and a Readiness summary.	GitHub API rate limits and account-level setup can remain warning/manual items until authenticated launch setup is complete.
python -B scripts/dev.py pr-triage	Prints Open PRs: 0 when no visible PRs await review, or lists each PR with risk findings and required gates.	If the API is rate-limited, public HTML fallback can prove no visible open PRs; authenticate before approving workflows or merging.

For recurring environment-specific failures, see Development Issue Solutions.

Troubleshooting pointers:

Situation	First Check
GitHub API rate limits or pending Actions status	Rerun python -B scripts/dev.py github-readiness after a short wait, or use an authenticated GitHub environment for account-level checks; see Development Issue Solutions.
Docker is unavailable locally	The verified default path is local Python. python -B scripts/dev.py container-release checks container files without Docker, while python -B scripts/dev.py docker-runtime is only for Docker-enabled machines; see Container Release Hygiene and Docker Runtime Evidence Checklist.
Optional OpenAI mode is unavailable	Local deterministic mode remains the default. python -B scripts/dev.py openai-live is an optional API-key-environment proof for model-facing routes only; see Model Runtime Configuration and OpenAI Live Mode Troubleshooting.
Generated local artifacts appear	Runtime outputs under ignored paths such as out/ are local evidence, not source changes. Run python -B scripts/dev.py safety before committing if a generated file appears in the worktree.

Release Evidence FAQ

Question	Answer
Which check should run before a local commit?	Use python -B scripts/dev.py quality; it proves local safety, docs/assets, UI contracts, service health, smoke flows, evals, replay artifacts, and claim checks are aligned.
When should fresh-clone-local run?	Run python -B scripts/dev.py fresh-clone-local before push when README, setup paths, public assets, or runtime wiring changed; it validates a clean clone of the current checkout before the remote has the commit.
What does remote fresh-clone prove?	After push, python -B scripts/dev.py fresh-clone clones origin, runs release-facing static gates, starts isolated services, and runs smoke flows from the public branch.
What does post-publish prove?	python -B scripts/post_publish_check.py proves the GitHub page, raw README/workflow, and required published files are reachable; compare with Published Repository Status.
Is a warning the same as a failing release gate?	No. Treat quality, fresh-clone-local, fresh-clone, and post-publish failures as blockers for code or docs changes; treat GitHub [WARN]/[MANUAL] items as account-level follow-up unless strict mode is being used.
How should GitHub readiness warnings be handled?	python -B scripts/dev.py github-readiness warnings for API rate limits, repository metadata, branch protection, release pages, social preview, or profile pin are account-level/manual items unless a gate reports a failure; see Development Issue Solutions.

Evidence Freshness Checklist

Ignored outputs under out/ are local evidence by default. Do not claim Docker runtime, OpenAI live mode, branch protection, or release page freshness until the matching command or account-level action is complete.

Evidence	Verify Or Refresh	Source Of Truth
README screenshots and mobile screenshots	Run python -B scripts/dev.py visual-assets and python -B scripts/dev.py visual-asset-diff; use python -B scripts/dev.py refresh-visual-assets only for intentional screenshot updates.	Visual Asset Hygiene and Screenshots.
Demo walkthrough GIF	Run python -B scripts/dev.py assets and inspect docs/assets/demo-walkthrough.gif before sharing; refresh it only for intentional demo-flow changes.	Visual Asset Hygiene.
Eval summary counts	Run python -B scripts/dev.py evals, python -B scripts/dev.py report, and python -B scripts/dev.py claims; commit source docs only when the claimed metrics intentionally change.	Demo Report and Evidence Matrix.
Published repository status	After push, run python -B scripts/dev.py fresh-clone, python -B scripts/post_publish_check.py, and python -B scripts/dev.py github-readiness; treat [WARN] and [MANUAL] rows as account-level follow-up unless strict mode is required.	Published Repository Status and Release Evidence FAQ.
Generated local artifacts	Run python -B scripts/dev.py replay-artifact when a fresh local replay package is needed; keep ignored out/ artifacts uncommitted unless a release process explicitly asks for them.	Demo Report and Release Evidence FAQ.

Current verified status:

health check: all services ok
smoke tests: 13/13 passed
Project 1 eval: 11/11 passed, unsafe_leak_failures = 0
Project 2 eval: 8/8 passed, unsafe_direct_side_effect_failures = 0
Project 3 eval: 6/6 passed, unsafe_release_approval_failures = 0

Demo Path Map

Project	Fastest Useful Path	Inspect Next
Secure Enterprise Knowledge Copilot	Open http://127.0.0.1:8765, select Alice, and ask What is the finance retention plan?; then switch to Morgan for the same question.	Compare abstention vs citation-backed access, copy the trace ID, then run python -B scripts/dev.py smoke.
Regulated Customer Operations Agent	Open http://127.0.0.1:8770, select Ivy and case-1001, then run the investigation.	Inspect the pending approval, blocked side effect, audit event, and python -B scripts/dev.py smoke.
AI Reliability Incident Console	Open http://127.0.0.1:8780, select the unsafe canary incident, then run triage.	Inspect failed eval evidence, blocked rollout, remediation steps, trace/audit records, and python -B scripts/dev.py smoke.

Demo-state reset presets:

docs/demo_state_presets.json stores the shareable local reset presets for the canonical Project 1 finance-access path, Project 2 case-1001 approval path, and Project 3 unsafe canary release path. Run python -B scripts/dev.py demo-presets to verify the preset IDs, reset commands, seed references, eval references, and POST payloads still match the fictional seed data before recording or sharing.

Demo recording readiness:

Use the Demo Recording Checklist with the Demo Path Map, Demo State Presets, Visual Asset Hygiene, and Demo Replay Artifact. Before recording, run python -B scripts/dev.py demo-presets, python -B scripts/dev.py smoke, python -B scripts/dev.py visual-assets, and python -B scripts/dev.py replay-artifact.

Launch-channel readiness:

Use the Launch Copy Pack with the Star Growth Plan, Launch Feedback Collection Examples, Stale Launch Feedback Claim Examples, Launch Assets Hygiene, GitHub Launch Plan, and Published Repository Status. Before sharing public launch posts, run python -B scripts/dev.py launch-assets, python -B scripts/dev.py assets, python -B scripts/dev.py fresh-clone, and python -B scripts/post_publish_check.py; keep Docker runtime, OpenAI live mode, branch protection, release pages, repo topics, profile pins, social preview setup, current feedback, stale feedback, and launch feedback as manual claims until their checks or account actions are complete.

Contribution safety readiness:

Use CONTRIBUTING, SECURITY, PR Review Security, Docs-Only PR Review Examples, Contributor Attribution Examples, Issue Triage SLA Wording Examples, Discussion To Issue Conversion Examples, PR Review Runbook, the Maintainer PR Checklist, and the Contributor Route Map before reviewing outside changes. For external PRs, inspect the diff before running untrusted code, then run python -B scripts/dev.py pr-triage, python -B scripts/dev.py pr-policy, python -B scripts/dev.py safety, and python -B scripts/dev.py quality; keep public credit tied to useful public GitHub activity or explicitly permissioned public attribution, keep issue replies separate from support SLAs or delivery guarantees, and keep discussion feedback separate from accepted issue scope.

Release artifact readiness:

Use the Demo Replay Artifact, Release Attachment Verification Examples, Release Asset Upload Dry-Run Examples, Release Asset Checksum Mismatch Examples, Release Note Refresh Checklist, Release Note Changelog Drift Examples, GitHub Release Attachment Screenshot Checklist, GitHub Release Commands, Post-Publish Checklist, Published Repository Status, and Final Readiness Report before preparing release evidence. Regenerate attachable evidence with python -B scripts/dev.py replay-artifact, then run python -B scripts/dev.py fresh-clone, python -B scripts/post_publish_check.py, and python -B scripts/dev.py quality; do not claim a GitHub release page, release notes, changelog summaries, release asset checksums, or release assets are current until the page exists, the public text matches the checked-in notes, and the artifacts are attached or linked.

Optional-environment readiness:

Use Container Release Hygiene, Model Runtime Configuration, Model Gateway Safety, OpenAI Live Mode Troubleshooting, the GitHub readiness notes in Published Repository Status, and Development Issue Solutions before claiming optional environment evidence. Run python -B scripts/dev.py container-release for static container config, python -B scripts/dev.py docker-runtime only on a Docker-enabled machine, python -B scripts/dev.py openai-live only in an API-key environment, and python -B scripts/dev.py github-readiness after a public push; Docker runtime and OpenAI live mode stay manual or environment-dependent until their matching commands pass in the right environment.

Connector roadmap readiness:

Use Production Upgrade Notes, Project Case Notes, Model Gateway Safety, Architecture Boundaries, and the Connector stubs row before adding external-system adapters. Connector stubs must keep external side effects behind approval, idempotency, audit, and trace boundaries; run python -B scripts/dev.py architecture, python -B scripts/dev.py model-gateway-safety, python -B scripts/dev.py contracts, and python -B scripts/dev.py quality.

Eval regression readiness:

Use Demo Report, Demo Replay Artifact, System Evidence Matrix, Scenario Data Integrity, Eval Authoring Guide, Eval CSV Troubleshooting Examples, and the Evidence Legend before changing eval or regression evidence. Run python -B scripts/dev.py evals, python -B scripts/dev.py eval-csv, python -B scripts/dev.py claims, and python -B scripts/dev.py quality; unsafe leak, unsafe direct side-effect, and unsafe release approval failure counts must remain zero.

Storage adapter readiness:

Use PostgreSQL And pgvector Adapter Design, Production Upgrade Notes, Scenario Data Integrity, Architecture Boundaries, and the Evidence Matrix before adding persistent storage prototypes. Storage adapters must preserve permission checks before retrieval or side effects, eval-state isolation, trace/audit compatibility, and local JSON default behavior; run python -B scripts/dev.py scenario-data, python -B scripts/dev.py dependency-surface, python -B scripts/dev.py contracts, and python -B scripts/dev.py quality.

Red-team eval readiness:

Use Threat Model, System Evidence Matrix, Scenario Data Integrity, Technical Review Playbook, and the Evidence Legend before adding attack or bypass eval cases. Prompt-injection, unauthorized-access, approval-bypass, and unsafe-release cases must keep unsafe leak, unsafe direct side-effect, and unsafe release approval failure counts at zero; run python -B scripts/dev.py threat-model, python -B scripts/dev.py scenario-data, python -B scripts/dev.py evals, and python -B scripts/dev.py quality.

OpenTelemetry backend readiness:

Use OpenTelemetry Trace Export, Observability Integrity, System Evidence Matrix, Production Upgrade Notes, and the Evidence Legend before adding hosted collector support. Local JSON trace export remains the default proof path and hosted collectors are optional environment-dependent extensions; run python -B scripts/dev.py replay, python -B scripts/dev.py otel-traces, python -B scripts/dev.py observability, and python -B scripts/dev.py quality.

Docker runtime readiness:

Use Container Release Hygiene, Docker Runtime Evidence Checklist, Production Upgrade Notes, Published Repository Status, System Evidence Matrix, and the Release Evidence FAQ before claiming container runtime evidence. python -B scripts/dev.py container-release is the local static proof path; python -B scripts/dev.py docker-runtime is environment-dependent and should be claimed only after it passes on a Docker-enabled machine. Run python -B scripts/dev.py container-release, python -B scripts/dev.py fresh-clone-local, and python -B scripts/dev.py quality before publishing Docker-facing docs.

Dependency surface readiness:

Use Supply Chain Security, Dependabot And Secret-Scanning Verification Examples, Stale Dependabot Alert Evidence Examples, System Evidence Matrix, the Contributor Route Map, Development Issue Solutions, and the Evidence Matrix before adding packages, CDNs, runtime manifests, or dependency-monitoring claims. The default posture is stdlib-only Python, first-party frontend assets, pinned Docker bases, and explicit Dependabot coverage until a dependency is intentionally reviewed; run python -B scripts/dev.py dependency-surface, python -B scripts/dev.py safety, python -B scripts/dev.py workflow-security, and python -B scripts/dev.py quality.

API contract readiness:

Use API Contracts, System Evidence Matrix, Architecture Boundaries, Runtime UI Contracts, and the Evidence Matrix before changing backend routes or frontend API calls. Documented response shapes, read-only scenario snapshot endpoints, static route handling, and frontend API modules must stay aligned; run python -B scripts/dev.py api-docs, python -B scripts/dev.py contracts, python -B scripts/dev.py ui-contracts, and python -B scripts/dev.py quality.

Error hygiene readiness:

Use Error Hygiene, System Evidence Matrix, Runtime UI Contracts, API Contracts, and the Evidence Matrix before changing app shells or API error handling. Unexpected server exceptions must return generic JSON errors without local paths, stack details, source file names, secret-like markers, or raw exception text, while typed application errors can still return user-safe messages; run python -B scripts/dev.py error-hygiene, python -B scripts/dev.py safety, python -B scripts/dev.py ui-contracts, and python -B scripts/dev.py quality.

Model gateway readiness:

Use Model Gateway Safety, Model Runtime Configuration, OpenAI Live Mode Troubleshooting, System Evidence Matrix, Production Upgrade Notes, and the Evidence Matrix before changing gateway code or runtime configuration. OpenAI mode stays opt-in, API keys stay outside the repo, structured outputs are required, failures fall back locally, and models do not authorize permissions, approvals, audit logs, or eval success; run python -B scripts/dev.py model-gateway-safety, python -B scripts/dev.py safety, python -B scripts/dev.py contracts, and python -B scripts/dev.py quality, with python -B scripts/dev.py openai-live only in API-key environments before claiming live model evidence.

PR triage readiness:

Use PR Review Security, Docs-Only PR Review Examples, PR Review Runbook, GitHub Public PR API Fallback Troubleshooting Examples, the Maintainer PR Checklist, Development Issue Solutions, and the Evidence Matrix before approving workflows, running contributor code, or merging public contributions. Treat public PRs as untrusted input: read changed files and diffs before running code, and give workflow, dependency, model gateway, safety, eval, binary, network, shell, and environment changes extra scrutiny; run python -B scripts/dev.py pr-triage, python -B scripts/dev.py pr-policy, python -B scripts/dev.py safety, and python -B scripts/dev.py quality.

Threat model readiness:

Use Threat Model, System Evidence Matrix, the Evidence Legend, Architecture Boundaries, and the Evidence Matrix before changing security-sensitive behavior. Information disclosure, prompt injection, unsafe side effects, public PR abuse, dependency drift, optional model gateway risk, observability gaps, and UI route surprises must map to deterministic owners and evidence commands; run python -B scripts/dev.py threat-model, python -B scripts/dev.py safety, python -B scripts/dev.py evals, and python -B scripts/dev.py quality.

Workflow security readiness:

Use Workflow Security, GitHub Actions Warning Examples, GitHub Actions Badge Verification Examples, Stale GitHub Actions Badge Cache Examples, System Evidence Matrix, PR Review Security, PR Review Runbook, and the GitHub Actions row in the Evidence Matrix before changing GitHub Actions, README badges, or automation paths. Public PR workflows must preserve read-only tokens, no secrets, safe triggers, hardened checkout, approved actions, and no CI push or GitHub authentication path; run python -B scripts/dev.py workflow-security, python -B scripts/dev.py governance, python -B scripts/dev.py pr-policy, and python -B scripts/dev.py quality.

Governance readiness:

Use GitHub Repository Settings, branch protection payload, Maintainer Review Policy, CODEOWNERS, and the repository governance row in the Evidence Matrix before changing maintainer, release, or contribution rules. Repository policy changes must preserve CODEOWNERS coverage, branch-protection expectations, issue and PR templates, security review rules, release evidence gates, and dependency monitoring; run python -B scripts/dev.py governance, python -B scripts/dev.py pr-policy, python -B scripts/dev.py workflow-security, and python -B scripts/dev.py quality.

Launch asset readiness:

Use Launch Assets Hygiene, Launch Copy Pack, Star Growth Plan, Launch Feedback Collection Examples, Stale Launch Feedback Claim Examples, GitHub Launch Plan, Published Repository Status, and the launch materials row in the Evidence Matrix before changing launch docs, posts, screenshots, feedback wording, or release-facing status. Public copy must keep Docker runtime, OpenAI live mode, branch protection, release pages, social preview, profile pins, launch feedback, stale feedback, star-growth success, and production readiness unclaimed until the matching evidence or account-level action exists; run python -B scripts/dev.py launch-assets, python -B scripts/dev.py assets, python -B scripts/dev.py fresh-clone-local, and python -B scripts/dev.py quality.

Reviewer handoff readiness:

Use Reviewer Perspective Checklist, Final Demo Runbook, Final Readiness Report, the Demo Path Map, the Evidence Matrix, and Published Repository Status before presenting, sharing, or reviewing the repository. Reviewer-facing claims must match current evidence, and post-push GitHub readiness warnings remain manual until the matching account-level action or public check is complete; run python -B scripts/dev.py quality, python -B scripts/dev.py fresh-clone-local, python -B scripts/dev.py visual-assets, and python -B scripts/dev.py launch-assets.

Post-publish readiness:

Use Post-Publish Checklist, Post-Publish Warning Examples, Published Repository Status, GitHub Repository Settings, GitHub Release Commands, the Release Evidence FAQ, and the published repository row in the Evidence Freshness Checklist before sharing the public branch or claiming published evidence. Remote clone results, published files, raw README/workflow reachability, and GitHub readiness warnings are separate from local quality evidence; after push, run python -B scripts/dev.py fresh-clone, python -B scripts/post_publish_check.py, python -B scripts/dev.py github-readiness, and python -B scripts/dev.py quality.

GitHub readiness:

Use Published Repository Status, GitHub Repository Settings, GitHub Repository Settings Screenshot Checklist, GitHub Authenticated Maintenance Troubleshooting Examples, GitHub API Rate-Limit Troubleshooting Examples, GitHub Repository Metadata Troubleshooting Examples, Social Preview Verification Examples, GitHub Profile Pin Verification Examples, GitHub Label Troubleshooting Examples, GitHub Release Commands, Development Issue Solutions, and the GitHub readiness row in the Evidence Freshness Checklist before changing repository metadata, topics, labels, branch protection, release pages, social preview, profile pins, or community issue state. Treat [WARN] and [MANUAL] output for rate limits, metadata, topics, labels, branch protection, release page, social preview, and profile pin as account-level or remote-freshness follow-up until github-readiness or authenticated maintenance confirms them; run python -B scripts/dev.py github-readiness, python -B scripts/dev.py github-maintenance, python -B scripts/dev.py github-community, and python -B scripts/dev.py quality.

Release page readiness:

Use the Demo Replay Artifact, GitHub Release Commands, Release Note Refresh Checklist, GitHub Release Page Troubleshooting Examples, GitHub Latest Release Troubleshooting Examples, GitHub Release Attachment Screenshot Checklist, Release Asset Checksum Mismatch Examples, Post-Publish Checklist, Published Repository Status, Final Readiness Report, and the Release Evidence FAQ before claiming the v0.1.0 GitHub release page is ready. The release page is remote evidence only after the page exists, the latest-release state points to the intended tag, the public page text matches docs/github_release_notes_v0.1.0.md, and current out/demo_replay_artifact.md and out/demo_replay_artifact.json outputs are attached or linked with reviewed checksum evidence; run python -B scripts/dev.py replay-artifact, python -B scripts/dev.py fresh-clone, python -B scripts/post_publish_check.py, python -B scripts/dev.py github-readiness, and python -B scripts/dev.py quality.

Operational runbook index:

Scenario	Fast Path	Evidence To Inspect
Project 1 retrieval, citation-backed answer, and unauthorized abstention	Use the Demo Path Map Alice/Morgan finance path and the Project 1 sequence in Final Demo Runbook.	Project Case Notes, Technical Review Playbook, and the permission-aware RAG rows in the Evidence Matrix.
Project 2 investigation, approval queue, side-effect blocking, and supervisor approval	Use the Demo Path Map Ivy case-1001 path and the Project 2 sequence in Final Demo Runbook.	Project Case Notes, Technical Review Playbook, and the governed tool-use rows in the Evidence Matrix.
Project 3 unsafe release triage, failed-eval evidence, rollout blocking, and remediation	Use the Demo Path Map unsafe canary path and the reliability-console review flow in Project Case Notes.	Final Demo Runbook, Technical Review Playbook, and the release-triage rows in the Evidence Matrix.

Core Terms

Term	Meaning In This Repository
Release gate	The repository-level checks that keep public docs, evidence, runtime contracts, screenshots, and safety claims aligned before a change is published; see the Evidence Matrix and launch asset hygiene.
Eval gate	Deterministic regression cases that must keep permission leaks, unsafe side effects, and unsafe release approvals at zero; see python -B scripts/dev.py evals and the System Evidence Matrix.
Approval gate	Application code that blocks external side effects until an authorized supervisor approves the pending action; see Project 2 and observability integrity.
Trace ID	A per-response identifier that connects UI output to stored trace records, linked audit events, approvals, blocked actions, or release decisions; see observability integrity.
Audit log	Structured records of security, workflow, approval, and release-decision events that explain what happened after a run; see threat model and observability integrity.
Abstention	The answer behavior used when accessible evidence is missing, unauthorized, or unsafe after filtering; see Project 1 and the System Evidence Matrix.

Evidence Legend

Gate	Proves	Does Not Prove
Smoke	python -B scripts/dev.py smoke proves the three running demos complete the canonical permission, approval, and release-blocking flows.	It is not exhaustive security, load, or browser-compatibility coverage.
Eval	python -B scripts/dev.py evals proves deterministic regression cases keep unsafe leak, direct side-effect, and release-approval failures at zero; see System Evidence Matrix.	It does not cover every possible prompt, data set, or production integration.
Trace	python -B scripts/dev.py observability proves responses can be followed through stored trace records, IDs, and linked decisions; see Observability Integrity.	It does not mean an external OpenTelemetry backend is configured by default.
Audit	The same observability gate proves security, approval, blocked-action, and release-decision events are recorded and link back to the run.	It does not replace enterprise retention, SIEM, or compliance controls.
Visual	python -B scripts/dev.py visual-assets proves desktop and mobile screenshots match the recorded manifest, source hashes, and contrast samples; see Visual Asset Hygiene.	It is a deterministic screenshot guard, not a complete accessibility audit.

Maintainer PR Checklist

Public PRs are treated as untrusted input. Before approving workflows, running contributor code, or merging:

Check	Action
Triage first	Run python -B scripts/dev.py pr-triage, then read the changed files and diff before running code.
High-risk surfaces	Treat workflow files, dependency policy, model gateways, safety scans, quality gates, shell commands, network calls, and binary/generated artifacts as high scrutiny.
Secrets and access	Do not ask contributors for secrets, tokens, account access, private files, local paths, or collaborator permissions.
Merge bar	Use the PR review security gate and PR review runbook; merge only after pr-policy, governance, workflow-security, safety, and verify pass.

Reviewer Handoff Checklist

Use this with the Maintainer PR Checklist, Contributor Route Map, Visual Asset Hygiene, Published Repository Status, PR Review Runbook, and Release Evidence FAQ.

Review Surface	Command Or Source	Blocking Rule
Local quality	Run python -B scripts/dev.py quality after source, docs, asset, or runtime changes.	Any failure blocks commit, approval, or public sharing.
Local fresh clone	Run python -B scripts/dev.py fresh-clone-local before push when setup paths, README guidance, public assets, or runtime wiring changed.	Any failure blocks push until a clean checkout works.
Visual evidence	Run python -B scripts/dev.py visual-assets and python -B scripts/dev.py visual-asset-diff; use refresh-visual-assets only for intentional screenshot updates.	Unexpected screenshot or manifest drift blocks visual claims.
Remote evidence	After push, run python -B scripts/dev.py fresh-clone and python -B scripts/post_publish_check.py.	Remote clone or published-file failures block sharing the public branch.
PR state	Run python -B scripts/dev.py pr-triage, then inspect high-risk diffs before running contributor code.	Do not merge while review findings, high-risk files, or required gates are unresolved.
GitHub readiness	Run python -B scripts/dev.py github-readiness; treat [WARN] and [MANUAL] rows as account-level follow-up unless strict mode is required.	Do not claim metadata, branch protection, release page, social preview, or profile pin freshness until those actions are complete.

Contributor Route Map

Change Area	Read First	Run Before PR
Docs-only	The command decision tree above, Launch Asset Hygiene, and System Evidence Matrix.	python -B scripts/dev.py assets, python -B scripts/dev.py launch-assets, then python -B scripts/dev.py quality before publishing.
Frontend/UI	Frontend Integrity, Runtime UI Contracts, and Visual Asset Hygiene.	python -B scripts/dev.py frontend, python -B scripts/dev.py ui-contracts, python -B scripts/dev.py visual-assets, then python -B scripts/dev.py quality.
Backend/API	API Contracts, Architecture Boundaries, and the service src/ package being changed.	python -B scripts/dev.py contracts, python -B scripts/dev.py api-docs, python -B scripts/dev.py architecture, then python -B scripts/dev.py quality.
Eval/data	System Evidence Matrix, Scenario Data Integrity, Eval Authoring Guide, Seed Data Extension Examples, and the project data/ folder.	python -B scripts/dev.py evals, python -B scripts/dev.py scenario-data, python -B scripts/dev.py claims, then python -B scripts/dev.py quality.
Visual assets	The Screenshots section and Visual Asset Hygiene.	python -B scripts/dev.py visual-assets and python -B scripts/dev.py visual-asset-diff; use python -B scripts/dev.py refresh-visual-assets only for intentional screenshot updates.
GitHub maintenance	Maintainer PR Checklist, PR Review Security, PR Review Runbook, and GitHub Authenticated Maintenance Troubleshooting Examples.	python -B scripts/dev.py pr-triage, python -B scripts/dev.py github-readiness, and dry-run python -B scripts/dev.py github-maintenance before any account-level action.

Production upgrade pointer:

Upgrade Path	Start With	Verification Boundary
FastAPI service adapter	Production Upgrade Notes, API Contracts, and Architecture Boundaries.	Keep the stdlib HTTP server as the default local path; run python -B scripts/dev.py contracts, python -B scripts/dev.py api-docs, and python -B scripts/dev.py quality.
PostgreSQL and pgvector	PostgreSQL And pgvector Adapter Design.	Preserve permission checks before retrieval or side effects, keep eval state isolated, and run python -B scripts/dev.py scenario-data plus python -B scripts/dev.py quality.
Connector stubs	Production Upgrade Notes and project service packages.	Keep external side effects behind approval, idempotency, audit, and trace boundaries; run python -B scripts/dev.py model-gateway-safety, python -B scripts/dev.py contracts, and python -B scripts/dev.py quality.
OpenTelemetry export	OpenTelemetry Trace Export and Observability Integrity.	Local traces export without a collector by default; run python -B scripts/dev.py replay, python -B scripts/dev.py otel-traces, and python -B scripts/dev.py observability.
OpenAI runtime mode	Model Runtime Configuration, Model Gateway Safety, and OpenAI Live Mode Troubleshooting.	Local deterministic mode remains the verified default; run python -B scripts/dev.py openai-live only in an API-key environment before claiming live model evidence.
Docker runtime	Container Release Hygiene and Docker Runtime Evidence Checklist.	Static container config is covered by python -B scripts/dev.py container-release; run python -B scripts/dev.py docker-runtime on a Docker-enabled machine before claiming container runtime evidence.

Evidence Matrix

Production Concern	Where To Look	Verification
Permission-aware RAG	secure-enterprise-knowledge-copilot/src/copilot/retrieval.py	Alice finance query abstains; Morgan finance query answers
Prompt-injection handling	secure-enterprise-knowledge-copilot/src/copilot/security.py, secure-enterprise-knowledge-copilot/src/copilot/answering.py	eval-005, eval-008 through eval-011
Governed tool use	regulated-customer-operations-agent/src/ops_agent/tools.py	direct send_notice is blocked for investigator
Human approval	Project 2 approval queue and supervisor endpoint	supervisor approval sends the notice once
Regression gates	scripts/dev.py, project eval runners, CSV summary export	python -B scripts/dev.py verify, python -B scripts/dev.py eval-csv
Public claim consistency	scripts/check_claim_consistency.py	README, release notes, evidence matrix, and preview metrics match eval/smoke evidence
Architecture boundaries	scripts/check_architecture_boundaries.py, docs/architecture_boundaries.md	app shells, API classes, backend packages, and frontend modules preserve separation of concerns
Frontend integrity	scripts/check_frontend_integrity.py, project web/ folders	HTML, labels, local ES modules, DOM wiring, trace-copy, trace-link, keyboard trace navigation, copyable scenario-draft controls, local draft diffs, accessibility CSS, browser-local theme controls, and quick actions stay intact
Visual asset hygiene	scripts/check_visual_asset_manifest.py, scripts/refresh_visual_assets.py, docs/visual_assets_manifest.json	Desktop and mobile demo screenshots stay tied to recorded frontend source hashes, contrast samples, and live local refreshes
Fresh clone experience	scripts/check_fresh_clone_experience.py, docs/fresh_clone_experience.md	clone the public repo into a temp directory, run release-facing checks, start all apps on isolated ports, and run smoke flows
Runtime UI contracts	scripts/check_runtime_ui_contracts.py, project app.py files	static assets, trace keyboard modules, scenario editor modules, content types, security headers, 404s, and traversal blocking
Error hygiene	scripts/check_error_hygiene.py, project app.py files	unexpected exceptions return generic JSON errors without leaking paths, stack details, or secret-like strings
Dependency surface	scripts/check_dependency_surface.py, .github/dependabot.yml, docs/supply_chain_security.md	stdlib-only Python path, first-party frontend assets, pinned Docker bases, and Dependabot coverage
Container release hygiene	scripts/check_container_release.py, scripts/check_docker_runtime.py, docs/container_release_hygiene.md	Dockerfiles, compose ports, health checks, startup commands, env handling, build-context ignores, and optional runtime smoke checks stay aligned
API contracts	scripts/check_api_contracts.py, scripts/check_api_documentation.py, docs/api_contracts.md	runtime response shapes and public API documentation stay aligned with source routes
GitHub launch setup	scripts/configure_github_launch.py	dry-run repo metadata, topics, branch protection, release commands, and release asset upload commands
Community issue pack	scripts/check_community_issue_pack.py, scripts/manage_community_issues.py, docs/github_labels.json	labels, issue templates, contributor issue pack, and optional GitHub issue creation stay aligned
Launch asset hygiene	scripts/check_launch_assets.py, docs/launch_assets_hygiene.md	launch copy, star-growth plan, public issue pack, and anti-hype boundaries stay complete and honest
Repository governance	scripts/check_repository_governance.py, .github/CODEOWNERS	code-owner review and branch-protection payload sanity checks
Workflow security	scripts/check_workflow_security.py, .github/workflows/ci.yml	read-only workflow token, safe PR trigger, hardened checkout, and approved actions
Model gateway safety	scripts/check_model_gateway_safety.py, scripts/check_openai_live_mode.py, project model_gateway.py files	OpenAI mode is opt-in, key references are constrained, structured outputs are required, failures fall back locally, and live mode can be verified when a key is available
Observability integrity	scripts/check_observability_integrity.py, project trace/audit/approval endpoints	response trace IDs, audit events, approval records, blocked actions, unauthorized-query evidence, and release decisions stay internally consistent
Threat model	docs/threat_model.md, scripts/check_threat_model.py	threat IDs map to deterministic controls, source files, and evidence commands
Scenario data integrity	scripts/check_scenario_data_integrity.py, project data/ folders	fictional seed data, roles, references, eval expectations, and browser-local scenario drafts remain internally consistent
Demo state presets	scripts/check_demo_state_presets.py, docs/demo_state_presets.json	shareable reset presets for the Project 1 finance-access path, Project 2 case-1001 approval path, and Project 3 unsafe canary release path stay aligned with seed and eval data
PR review policy	scripts/check_pr_review_policy.py, docs/pr_review_security.md	triage heuristics, runbook, maintainer policy, and PR template keep malicious-contribution checks intact
Public PR triage	scripts/review_open_prs.py	inspect open PRs and flag risky diffs before running code
Replayable demo	scripts/replay_demo.py	reset services, run key flows, print trace and approval evidence
Replay artifact	scripts/export_demo_replay_artifact.py, docs/demo_replay_artifact.md	generate release-attachable Markdown and JSON evidence under ignored out/
Observability export	scripts/export_traces_otel.py	local traces convert to OTLP/JSON-compatible resourceSpans
Final readiness report	scripts/generate_final_readiness_report.py	compact launch, blocker, and technical review walkthrough status

See System Evidence Matrix for the full claim-to-evidence map.

Screenshots

Secure Enterprise Knowledge Copilot	Regulated Customer Operations Agent	AI Reliability Incident Console
Desktop: role-aware knowledge access, visible documents, eval gate, and trace/audit surfaces for permission-aware RAG.	Desktop: investigator workflow with case context, governed action buttons, eval gate, and approval-driven operations controls.	Desktop: release and incident triage workspace with eval evidence, rollout blocking, and audit/trace context.

Mobile / narrow viewport captures are checked by the same visual asset manifest:

Secure Enterprise Knowledge Copilot	Regulated Customer Operations Agent	AI Reliability Incident Console
Mobile: narrow layout keeps user context, visible documents, and permission-aware knowledge controls readable.	Mobile: approval workflow remains usable with case selection, eval gate, and governed action controls stacked for scanning.	Mobile: release gate and incident triage stay readable while preserving blocked-rollout evidence.

Screenshot reviewer checklist:

Check	Evidence
Desktop and mobile assets cover all three demos.	Six PNGs are listed above and checked by python -B scripts/dev.py visual-assets.
The screenshots still match the live behavior.	Run python -B scripts/dev.py start, follow the Demo Path Map, and compare the visible role, approval, release, trace, and audit surfaces.
Refreshed screenshots are reviewable.	Run python -B scripts/dev.py visual-asset-diff and keep refreshed PNGs plus docs/visual_assets_manifest.json in the same change.
Reviewer expectations stay honest.	Use Visual Asset Hygiene and the Reviewer Perspective Checklist before publishing or approving visual changes.

Project 1: Secure Enterprise Knowledge Copilot

Open:

http://127.0.0.1:8765

Show:

1. Alice asks: How many days per week can employees work remotely?
2. The system answers with Remote Work Policy 2026 citation.
3. Alice asks: What is the finance retention plan?
4. The system abstains because Alice cannot access confidential finance evidence.
5. Morgan asks the same finance question.
6. The system answers with Finance Retention Plan 2026 citation.
7. Run evals.

Core claim:

The model never receives evidence the user is not allowed to access.

Project 2: Regulated Customer Operations Agent

Open:

http://127.0.0.1:8770

Show:

1. Ivy investigates Market Blue / RX-900 recalled product.
2. The agent searches policy and listings.
3. It creates a violation, drafts seller notice, schedules follow-up.
4. It creates an approval request before sending.
5. Direct send_notice is blocked for investigator.
6. Supervisor approval sends the notice.
7. Run evals.

Core claim:

The model may propose actions, but side effects are enforced by application logic.

Architecture

Reference Systems
  +-- Secure Enterprise Knowledge Copilot
  |   +-- role-aware retrieval
  |   +-- citation answer shape
  |   +-- abstention logic
  |   +-- prompt-injection detection
  |   +-- trace/audit/evals
  |
  +-- Regulated Customer Operations Agent
  |   +-- intent routing
  |   +-- business tools
  |   +-- approval queue
  |   +-- side-effect blocking
  |   +-- trace/audit/evals
  |
  +-- AI Reliability Incident Console
      +-- canary release triage
      +-- eval regression evidence
      +-- rollout blocking
      +-- remediation plans
      +-- trace/audit/evals

All three projects are dependency-free by default so they run reliably anywhere with Python 3.12. Optional OpenAI gateways are included for the model-facing projects but disabled by default.

Optional OpenAI Mode

$env:OPENAI_API_KEY="..."
$env:OPENAI_MODEL="gpt-5.2"
$env:OPENAI_REASONING_EFFORT="medium"
$env:OPENAI_TEXT_VERBOSITY="low"
$env:COPILOT_MODEL_PROVIDER="openai"
$env:OPS_AGENT_MODEL_ROUTER="openai"

Security boundaries remain outside the model:

• Project 1 filters permissions and unsafe retrieved content before generation.
• Project 2 enforces approval gates in deterministic application code.
• Project 3 links eval regressions to release decisions without a model dependency.

With a real API key, verify live OpenAI mode without changing the default local path:

python -B scripts/dev.py openai-live

The check starts both model-facing apps with OpenAI mode enabled, requires Project 1 to return model_provider=openai, requires Project 2 to return model_router=openai, and still verifies citations, approvals, and side-effect blocking.

Docker

Docker config is included:

docker compose up --build

Docker release hygiene is statically gated:

python -B scripts/dev.py container-release

On a Docker-enabled machine, run the runtime proof:

python -B scripts/dev.py docker-runtime

This builds all service images through Compose, waits for health endpoints, runs the same smoke flows against the containers, and tears the Compose project down. Docker is not installed in this environment, so the local Python runtime is the fully verified path here.

Repository Structure

repository/
  secure-enterprise-knowledge-copilot/
  regulated-customer-operations-agent/
  ai-reliability-incident-console/
  scripts/
  docs/
  .github/

Key Docs

• Project Content Index
• Changelog
• Final Demo Runbook
• Demo Report
• Demo Replay Artifact
• Project Case Notes
• Production Upgrade Notes
• PostgreSQL And pgvector Adapter Design
• OpenTelemetry Trace Export
• OpenTelemetry Collector Handoff Troubleshooting
• Model Runtime Configuration
• Model Gateway Safety
• OpenAI Live Mode Troubleshooting
• Launch Assets Hygiene
• Observability Integrity
• Trace Timeline Examples
• Threat Model
• Scenario Data Integrity
• Seed Fixture Data Flow
• Local Artifact Glossary
• Local Demo Reset Troubleshooting
• Command Output Troubleshooting Map
• Eval Authoring Guide
• Eval Gate Troubleshooting Examples
• Eval CSV Troubleshooting Examples
• Seed Data Extension Examples
• Error Hygiene
• Container Release Hygiene
• Docker Runtime Evidence Checklist
• Docker Runtime Failure Examples
• Visual Asset Hygiene
• Architecture Boundaries
• Contributor Code Tour
• First Pull Request Checklist
• Issue To PR Handoff Flow
• Contributor Attribution Examples
• Issue Triage SLA Wording Examples
• Discussion To Issue Conversion Examples
• Public Roadmap Issue Comment Examples
• Roadmap Duplicate Issue Handling Examples
• Issue Template Stale Evidence Examples
• README Navigation Audit
• README Navigation Drift Examples
• Workflow Security
• Dependabot And Secret-Scanning Verification Examples
• Stale Dependabot Alert Evidence Examples
• GitHub Actions Warning Examples
• GitHub Actions Badge Verification Examples
• Stale GitHub Actions Badge Cache Examples
• Final Completion Audit
• GitHub Launch Plan
• Published Repository Status
• Public Maintainer Status Update Examples
• Post-Publish Warning Examples
• GitHub Discussions Launch Checklist
• Stale GitHub Discussions Pin Examples
• GitHub Repository Settings
• GitHub Repository Settings Screenshot Checklist
• Stale Repository Topics Evidence Examples
• Stale Social-Preview Cache Examples
• Stale Profile-Pin Evidence Examples
• GitHub Label Troubleshooting Examples
• Branch Protection Verification Examples
• Stale Branch-Protection Screenshot Examples
• GitHub Release Page Troubleshooting Examples
• Community Backlog
• Public Release Audit
• Differentiation Strategy
• Technical Review Playbook
• System Design Deep Dive
• System Evidence Matrix
• ADR 0001: Local-First Reference Runtime
• ADR 0002: The Model Is Not The Security Boundary
• ADR 0003: Eval State Isolated From Demo State
• Secure RAG Case Study
• Governed Agent Case Study
• Demo Video Script
• Demo Recording Checklist
• Release Attachment Verification Examples
• Release Asset Upload Dry-Run Examples
• Release Asset Checksum Mismatch Examples
• Release Note Refresh Checklist
• Release Note Changelog Drift Examples
• GitHub Release Attachment Screenshot Checklist
• GitHub Release Page Troubleshooting Examples
• Stale Release-Page Screenshot Examples
• Star Growth Plan
• Launch Copy Pack
• Launch Feedback Collection Examples
• Stale Launch Feedback Claim Examples
• GitHub Community Issue Pack
• Reviewer Perspective Checklist
• Fresh Clone Experience
• API Contracts
• API Request Cookbook
• API Error Examples
• PR Review Security
• Docs-Only PR Review Examples
• Docs-Only Review Comment Examples
• Public Roadmap Issue Comment Examples
• Pull Request Review Runbook
• GitHub Release Commands
• Post-Publish Checklist
• Roadmap
• AI Reliability Incident Console

System Narrative

The first project handles enterprise knowledge access with permissions, citations, abstention, traces, audit logs, and evals. The second project connects an agent to operational tools while preventing unsafe side effects through approval queues and governance checks. The third project handles AI release reliability after deployment by linking eval regressions, incidents, rollout blocking, traces, audit logs, and remediation plans. The browser demos expose trace IDs, keyboard-friendly local trace links, visible focus states, reduced-motion CSS, browser-local light/dark theme controls, and copyable browser-local scenario drafts with local diffs so a specific run and its fictional seed context can be inspected again without relying on screenshots.

Together they form a compact reference architecture for AI systems that are useful, inspectable, and safe enough to reason about before enterprise deployment.

License

This project is released under the MIT License. See LICENSE.