@raishin/vanguard-frontier-agentic 1.0.0

package/README.md

@@ -0,0 +1,610 @@
+# 🚀 Vanguard Frontier Agentic
+
+**A friendly, curated marketplace for cloud and zero-trust AI workflows.**
+
+This repo collects reusable **skills**, **agents**, **rules**, **MCP references**,
+and supporting assets for engineers working with AWS, Azure, OCI, GCP,
+Kubernetes, Terraform, cloud security, and compliance-heavy architecture.
+
+Think of it as a toolbox:
+
+- 🧠 **Skills** = step-by-step workflows an AI assistant can follow.
+- 🤖 **Agents** = reusable expert roles for review, architecture, and operations.
+- 📏 **Rules** = durable instructions for a specific AI harness.
+- 🔌 **MCP references** = trusted notes for connecting tools to real systems.
+- 🗂️ **Catalogs** = machine-readable indexes so tools can discover everything.
+
+The goal is simple: **make AI-assisted cloud work safer, reusable,
+compliance-aware, and easier to understand.**
+
+> 📦 **npm status (verified 2026-04-28):** `@raishin/vanguard-frontier-agentic`
+> is **not published yet** on the public npm registry. Check live status with:
+> `npm view @raishin/vanguard-frontier-agentic version`
+
+---
+
+## 🌍 Vision
+
+**Build a practical AI workflow marketplace for secure cloud engineering.**
+
+This repository exists for teams that need to design, review, and operate cloud
+systems where security and compliance are not optional extras.
+
+The north star:
+
+> 🛡️ **Cloud architecture should be zero-trust by default, evidence-backed by
+> design, and understandable by engineers of any seniority.**
+
+That means every serious workflow should help engineers answer:
+
+- 👤 **Who is accessing what?**
+- 🔐 **Why are they allowed?**
+- 🧾 **Where is the evidence?**
+- 🚨 **How do we detect abuse or drift?**
+- 🧯 **How do we respond and recover?**
+- 📋 **Which compliance obligation does this support?**
+
+---
+
+## 🧬 Philosophy
+
+This repo is opinionated. That is a feature, not a bug.
+
+### 1. 🛡️ Zero trust beats implicit trust
+
+Do not trust a network, cloud account, CI runner, agent, workload, or human just
+because it is "inside" something.
+
+Good assets should push for:
+
+- strong identity,
+- least privilege,
+- explicit authorization,
+- segmentation,
+- continuous verification,
+- logging and detection,
+- short-lived credentials where possible,
+- safe rollback paths.
+
+### 2. 🧾 Compliance needs evidence, not vibes
+
+SOC 2 Type 2, PCI DSS, NIS2, and NIST-style control frameworks are not passed by
+good intentions. They require repeatable controls and evidence over time.
+
+Good assets should produce or point to evidence:
+
+- policy decisions,
+- access reviews,
+- architecture diagrams,
+- ticket approvals,
+- logs and alerts,
+- backup and restore tests,
+- vulnerability and patch records,
+- incident response records,
+- change history.
+
+### 3. 🔐 Least privilege is the default
+
+If a workflow recommends broad admin access, it must explain why.
+
+If it cannot explain why, it should not recommend it.
+
+### 4. 🧪 Every claim needs a source or a validation path
+
+Cloud behavior changes. Compliance expectations evolve. Vendor services drift.
+
+So assets should clearly separate:
+
+- ✅ verified facts,
+- 🧠 engineering judgment,
+- ⚠️ assumptions,
+- ❓ unknowns.
+
+### 5. 🧯 Automation must have brakes
+
+AI-assisted automation should not become a fast path to production damage.
+
+Dangerous actions need:
+
+- read-only discovery first,
+- explicit approval,
+- scoped credentials,
+- dry-run or plan mode where possible,
+- rollback notes,
+- post-change validation.
+
+---
+
+## 📋 Compliance compass
+
+This repository is not a compliance product and does not replace auditors,
+QSAs, legal counsel, or official standards.
+
+It is a **control-aware engineering toolbox**. The assets should help teams
+design and collect evidence for common security expectations across frameworks.
+
+| Framework / standard              | What it pushes us to remember                                                                                                                                     | Repo design implication                                                                                   |
+| --------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------- |
+| 🔵 **SOC 2 Type 2**               | Controls must operate over a period of time, especially around security, availability, confidentiality, processing integrity, and privacy trust service criteria. | Workflows should leave evidence trails, not just one-time fixes.                                          |
+| 💳 **PCI DSS**                    | Cardholder data environments need scoped controls, secure configuration, access control, monitoring, vulnerability management, and testing.                       | Workflows should reduce scope, avoid broad access, and flag payment-data risk.                            |
+| 🇪🇺 **NIS2**                       | EU cybersecurity rules emphasize governance, risk management, incident reporting, supply-chain security, and management accountability.                           | Workflows should make ownership, reporting, and supplier/cloud dependencies explicit.                     |
+| 🧭 **NIST CSF 2.0**               | Cybersecurity risk management spans Govern, Identify, Protect, Detect, Respond, and Recover.                                                                      | Assets should not stop at prevention; they should include detection, response, and recovery.              |
+| 🏛️ **NIST SP 800-207 Zero Trust** | Access should be continuously evaluated and should not rely on implicit network trust.                                                                            | Agents and skills should challenge flat networks, permanent credentials, and unverified trust boundaries. |
+
+Ruthless correction: **NIS2** is the European cybersecurity directive. **NIST**
+is a U.S. standards body. If someone says “NIST2 European compliance,” they
+probably mean **NIS2** or they are mixing two different things.
+
+---
+
+## 🏗️ Architecture principles
+
+Use these principles when creating or reviewing assets:
+
+| Principle                   | What good looks like                                                                  |
+| --------------------------- | ------------------------------------------------------------------------------------- |
+| 👤 Identity-first           | Humans, workloads, agents, and CI/CD jobs have explicit identities.                   |
+| 🔐 Least privilege          | Permissions are narrow, justified, and reviewable.                                    |
+| 🧱 Segmented blast radius   | Network, account, project, subscription, tenancy, and data boundaries are deliberate. |
+| 🧾 Evidence by design       | The workflow naturally produces logs, approvals, diffs, plans, or reports.            |
+| 🔎 Continuous monitoring    | Detection is part of the design, not an afterthought.                                 |
+| 🧯 Recoverability           | Backups, restore tests, rollback, and incident response are considered upfront.       |
+| 🧭 Source-grounded guidance | Official docs and live state beat memory and assumptions.                             |
+| 🤝 Human accountability     | AI can assist, but owners still approve risk.                                         |
+
+---
+
+## 🧭 Quick map
+
+| Folder                     | What lives here                                  | Easy memory hook                       |
+| -------------------------- | ------------------------------------------------ | -------------------------------------- |
+| [`skills/`](skills/)       | Reusable workflows grouped by provider or domain | 🧠 “How do I do this task?”            |
+| [`agents/`](agents/)       | Expert roles grouped by provider or domain       | 🤖 “Who should review this?”           |
+| [`rules/`](rules/)         | Harness-specific instructions                    | 📏 “What behavior is always expected?” |
+| [`mcp/`](mcp/)             | MCP server references and trust notes            | 🔌 “What can this connect to?”         |
+| [`catalog/`](catalog/)     | JSON indexes for marketplace discovery           | 🗂️ “What assets exist?”                |
+| [`schemas/`](schemas/)     | Metadata validation contracts                    | ✅ “What fields are required?”         |
+| [`templates/`](templates/) | Starter templates for new assets                 | 🧱 “How do I add one?”                 |
+| [`docs/`](docs/)           | Quality rules, taxonomy, and marketplace notes   | 📚 “How should this repo work?”        |
+| [`assets/`](assets/)       | Logos and visual assets                          | 🎨 “What images can docs use?”         |
+
+---
+
+## 📦 Consumer install and export selected agents
+
+This repository is designed so consumers can install **selected marketplace
+agents** into their own project instead of copying everything by hand.
+
+### Current package status
+
+As of **2026-04-28**, the public npm package:
+
+```bash
+@raishin/vanguard-frontier-agentic
+```
+
+was verified as **not yet published** on npm.
+
+Live check:
+
+```bash
+npm view @raishin/vanguard-frontier-agentic version
+```
+
+If that command returns `404 Not Found`, the package is still unpublished.
+
+### Use it today from GitHub
+
+Until npm publishing is live, install from GitHub:
+
+```bash
+npm install github:Raishin/vanguard-frontier-agentic
+```
+
+### Export selected agents into a consumer repository
+
+After installation, this package ships a CLI:
+
+```bash
+vfa-export-agents
+```
+
+It copies selected agent harness files from this marketplace into the correct
+runtime folders in a consumer repository.
+
+List available agent IDs:
+
+```bash
+npx vfa-export-agents --list
+```
+
+Export one agent to **Claude Code**:
+
+```bash
+npx vfa-export-agents \
+  --platform claude-code \
+  --agents azure-cosmosdb-platform-operator-agent \
+  --repo /path/to/consumer-repo
+```
+
+Export one agent to **GitHub Copilot**:
+
+```bash
+npx vfa-export-agents \
+  --platform copilot \
+  --agents azure-cosmosdb-platform-operator-agent \
+  --repo /path/to/consumer-repo
+```
+
+Export one agent to **Kiro** (both IDE + CLI adapters):
+
+```bash
+npx vfa-export-agents \
+  --platform kiro \
+  --agents azure-cosmosdb-platform-operator-agent \
+  --repo /path/to/consumer-repo
+```
+
+Export **all** agents for a platform:
+
+```bash
+npx vfa-export-agents --platform codex --all --repo /path/to/consumer-repo
+```
+
+Overwrite existing exported files intentionally:
+
+```bash
+npx vfa-export-agents --platform copilot --all --repo /path/to/consumer-repo --force
+```
+
+### Platform destination folders
+
+The exporter writes into the destination repository using platform-native
+runtime paths:
+
+| Platform | Destination path(s) |
+| -------- | ------------------- |
+| Codex | `.codex/agents/` |
+| Claude Code | `.claude/agents/` |
+| GitHub Copilot | `.github/agents/` |
+| Cursor | `.cursor/agents/` |
+| Gemini CLI | `.gemini/agents/` |
+| Kiro IDE | `.kiro/agents/` |
+| Kiro CLI | `.kiro/agents/` |
+
+### Important limitation
+
+This exporter installs **custom agent files**, not full repo-level guidance.
+
+If the consumer also wants the repository-level instruction layer, they should
+use the matching project entrypoints in their own repo as appropriate:
+
+- `AGENTS.md`
+- `CLAUDE.md`
+- `GEMINI.md`
+- `.github/copilot-instructions.md`
+
+See [`docs/normalized-platform-matrix.md`](docs/normalized-platform-matrix.md)
+for the exact distinction between repo guidance and custom agent installation.
+
+---
+
+## 🧠 Skills
+
+Skills are practical workflows. They should help an engineer complete a task
+with less guesswork.
+
+Current provider layout:
+
+```text
+skills/
+├── aws/
+├── azure/
+├── gcp/
+├── kubernetes/
+├── oci/
+├── security/
+└── terraform/
+```
+
+Examples:
+
+- 🔐 [`skills/aws/aws-iam-least-privilege-review`](skills/aws/aws-iam-least-privilege-review/)  
+  Review AWS IAM policies and reduce unnecessary access.
+
+- 🟦 [`skills/azure/azure-rbac-review`](skills/azure/azure-rbac-review/)  
+  Review Azure RBAC assignments, scopes, and custom roles.
+
+- 🟥 [`skills/oci/oci-autonomous-database-architect`](skills/oci/oci-autonomous-database-architect/)  
+  Design and review Oracle Autonomous Database across OCI and multicloud options.
+
+- 🏗️ [`skills/oci/oci-exadata-database-architect`](skills/oci/oci-exadata-database-architect/)  
+  Stress-test Exadata architecture, migration, HA/DR, and deployment choices.
+
+- 🔌 [`skills/oci/oracle-oci-mcp-grounded-advisor`](skills/oci/oracle-oci-mcp-grounded-advisor/)  
+  Ground Oracle/OCI MCP and cloud advice in official sources.
+
+Rule of thumb: if the asset teaches **how to do a repeatable task**, it is
+probably a skill.
+
+---
+
+## 🤖 Agents
+
+Agents are reusable expert roles. They are organized by cloud provider or domain
+so engineers can find the right reviewer quickly.
+
+```text
+agents/
+├── aws/
+├── azure/
+├── gcp/
+├── oci/
+├── multi-cloud/
+├── security/
+└── terraform/
+```
+
+Example:
+
+- 🧱 [`agents/terraform/terraform-reviewer`](agents/terraform/terraform-reviewer/)  
+  Review Terraform modules, plans, provider usage, and state assumptions.
+
+Use an agent when you need a **role with judgment**, not just a checklist.
+
+---
+
+## 🔌 MCP references
+
+MCP references describe tool/server integrations and their trust boundaries.
+
+Examples:
+
+- 🟧 [`mcp/official/aws-mcp-servers.md`](mcp/official/aws-mcp-servers.md)
+- 🟦 [`mcp/official/azure-mcp-server.md`](mcp/official/azure-mcp-server.md)
+- 🟥 [`mcp/official/oracle-mcp-servers.md`](mcp/official/oracle-mcp-servers.md)
+
+Important: MCP tools may read or mutate real infrastructure. Treat them like
+production access, not like harmless documentation links.
+
+---
+
+## 🎨 Logos and assets
+
+Visual assets are organized for future documentation, marketplace cards, and
+generated outputs.
+
+```text
+assets/logos/
+├── cloud/
+│   └── oci/
+└── vendors/
+    └── oracle/
+```
+
+Use:
+
+- `cloud/<provider>/` for platform or service logos.
+- `vendors/<vendor>/` for company/vendor logos.
+- SVG first, PNG fallback when needed.
+
+See [`assets/logos/README.md`](assets/logos/README.md).
+
+---
+
+## ✅ Quality bar
+
+This repo is **not** a prompt junk drawer.
+
+Every cataloged asset should be:
+
+- 🔎 **Traceable** — includes official docs or clear provenance.
+- 🔐 **Security-aware** — explains access, risk, and least-privilege concerns.
+- 🧪 **Validated** — passes repo checks before being shared.
+- 🧭 **Scoped** — clearly says which provider, domain, and harness it supports.
+- 🧯 **Safe by default** — read-only discovery before mutation; approval before
+  dangerous actions.
+
+Hard no:
+
+- ❌ Secrets or credentials.
+- ❌ Vague “do everything” prompts.
+- ❌ Unsafe production mutation recipes.
+- ❌ Cloud claims with no source or verification path.
+
+For the detailed standard, read [`docs/quality-bar.md`](docs/quality-bar.md).
+
+---
+
+## 🗂️ Metadata contract
+
+Every cataloged asset needs metadata so people and tools can understand it.
+
+Required common fields:
+
+- `id`
+- `name`
+- `type`: `skill`, `agent`, `rule`, or `mcp-reference`
+- `provider`: `aws`, `azure`, `oracle`, `oci`, `gcp`, `kubernetes`,
+  `terraform`, `multi-cloud`, or `generic`
+- `harnesses`: one or more of `codex`, `copilot`, `claude-code`, `cursor`,
+  `gemini`, `kiro`, `other`
+- `summary`
+- `source_type`: `original`, `adapted`, or `reference-only`
+- `official_docs`
+- `security_notes`
+- `last_verified`
+- `path`
+
+MCP references also need:
+
+- official project/source URL
+- vendor
+- auth model
+- install/config example
+- unofficial/community warning when relevant
+
+---
+
+## 🔏 Skill integrity manifests
+
+Skills are executable guidance. Treat them like supply-chain artifacts.
+
+This repo uses:
+
+```text
+catalog/skill-manifest.json
+```
+
+to record SHA-256 hashes for every file under every cataloged skill directory.
+
+Why it matters:
+
+- 🧾 proves what changed between releases,
+- 🚨 catches accidental or unauthorized edits,
+- 📦 gives npm consumers a package-level integrity map,
+- 🧪 makes skill integrity testable in CI,
+- 🔍 supports audit evidence for security-conscious teams.
+
+After intentional skill edits, regenerate the manifest:
+
+```bash
+npm run manifest:write
+```
+
+Before release or review, check it:
+
+```bash
+npm run manifest:check
+```
+
+Ruthless truth: a manifest proves file integrity. It does **not** prove a skill
+is safe, correct, compliant, or officially endorsed. You still need review.
+
+---
+
+## 🧪 Validate your changes
+
+Before contributing or sharing changes, run:
+
+```bash
+npm run validate
+```
+
+Equivalent manual commands:
+
+```bash
+python tests/validate-catalog.py
+python tests/validate-skill-manifest.py
+python tests/validate-links.py --offline
+```
+
+Before a release, also run the online link check and npm package preview:
+
+```bash
+python tests/validate-links.py
+npm pack --dry-run
+```
+
+If validation fails, fix that first. A broken catalog makes the marketplace
+harder to trust.
+
+---
+
+## 📦 npm publishing and semantic versioning
+
+This repository is npm-ready through [`package.json`](package.json).
+
+Use SemVer: `MAJOR.MINOR.PATCH`.
+
+| Version bump | Use when                                                                                                                                 | Example           |
+| ------------ | ---------------------------------------------------------------------------------------------------------------------------------------- | ----------------- |
+| 🩹 `PATCH`   | Safe fixes: typos, metadata corrections, non-behavioral doc cleanup, manifest refresh after safe edits.                                  | `0.1.0` → `0.1.1` |
+| ✨ `MINOR`   | Backwards-compatible additions: new skills, new agents, new provider folders, new optional metadata.                                     | `0.1.0` → `0.2.0` |
+| 💥 `MAJOR`   | Breaking changes: removed/renamed IDs, moved paths without aliases, schema-required-field changes, incompatible manifest format changes. | `1.4.2` → `2.0.0` |
+
+While the package is below `1.0.0`, be extra explicit in release notes because
+minor bumps may still include early breaking design changes.
+
+Read the full policy in [`docs/release-versioning.md`](docs/release-versioning.md).
+
+---
+
+## 📚 Source anchors
+
+Use official sources when writing security or compliance-sensitive assets:
+
+- 🏛️ [NIST SP 800-207 Zero Trust Architecture](https://csrc.nist.gov/pubs/sp/800/207/final)
+- 🧭 [NIST Cybersecurity Framework](https://www.nist.gov/cyberframework)
+- 🇪🇺 [European Commission: NIS2 Directive](https://digital-strategy.ec.europa.eu/en/policies/nis2-directive)
+- 🇪🇺 [ENISA: NIS Directive 2](https://www.enisa.europa.eu/topics/state-of-cybersecurity-in-the-eu/cybersecurity-policies/nis-directive-2)
+- 💳 [PCI Security Standards Council Document Library](https://www.pcisecuritystandards.org/document_library)
+- 🔵 [AICPA SOC 2 Trust Services Criteria](https://www.aicpa-cima.com/topic/audit-assurance/audit-and-assurance-greater-than-soc-2)
+
+Prefer these over blog posts. Blog posts can help explain, but they are not the
+source of truth.
+
+---
+
+## 🧑‍💻 How to add a new asset
+
+1. 🧭 Pick the right folder.
+   - Cloud workflow? Put it under `skills/<provider>/`.
+   - Expert role? Put it under `agents/<provider-or-domain>/`.
+   - Harness instruction? Put it under `rules/<harness>/`.
+   - MCP info? Put it under `mcp/official/` or `mcp/community/`.
+
+2. 🧱 Start from a template.
+   - Use [`templates/skill-template`](templates/skill-template/) for skills.
+   - Use [`templates/agent-template`](templates/agent-template/) for agents.
+
+3. 🗂️ Add or update catalog metadata.
+   - Skills go in [`catalog/skills.json`](catalog/skills.json).
+   - Agents go in [`catalog/agents.json`](catalog/agents.json).
+   - Rules go in [`catalog/rules.json`](catalog/rules.json).
+   - MCP references go in [`catalog/mcp-references.json`](catalog/mcp-references.json).
+
+4. ✅ Run validation.
+
+5. 🧯 Check safety.
+   - No secrets.
+   - No broad permissions unless justified.
+   - No destructive actions without approval gates and rollback notes.
+
+---
+
+## 🛡️ Contribution stance
+
+The default answer to low-trust contributions is **no**.
+
+That is intentional. Cloud automation can break real systems.
+
+Good contributions are:
+
+- useful,
+- specific,
+- auditable,
+- source-backed,
+- safe by default,
+- friendly for engineers of any seniority.
+
+See:
+
+- [`CONTRIBUTING.md`](CONTRIBUTING.md)
+- [`SECURITY.md`](SECURITY.md)
+- [`docs/taxonomy.md`](docs/taxonomy.md)
+- [`docs/compatibility.md`](docs/compatibility.md)
+- [`docs/marketplace-model.md`](docs/marketplace-model.md)
+
+---
+
+## 🧠 Remember this
+
+```text
+Skills = workflows 🧠
+Agents = expert roles 🤖
+Rules = always-on behavior 📏
+MCP = real tool connections 🔌
+Catalog = searchable index 🗂️
+Validation = trust ✅
+```

package/agents/AGENTS.md

@@ -0,0 +1,21 @@
+# AGENTS.md
+
+## Purpose
+- Store reusable expert roles grouped by provider or domain.
+- Keep role definitions judgment-oriented; use skills for task procedures.
+
+## Patterns
+- `agents/<provider-or-domain>/<agent-id>/AGENT.md` is the role prompt.
+- `agents/<provider-or-domain>/<agent-id>/metadata.json` mirrors catalog fields.
+- `agents/<provider-or-domain>/<agent-id>/harnesses/` stores harness-specific variants.
+- `catalog/agents.json` path must match the agent folder.
+- Provider folders: `aws/`, `azure/`, `gcp/`, `oci/`, `multi-cloud/`, `security/`, `terraform/`.
+
+## Do Not Miss
+- Move agents by updating metadata path and `catalog/agents.json` in the same change.
+- Do not leave empty placeholder agents in the catalog.
+- Do not flatten harness variants into provider roots; keep canonical identity first.
+- Run `npm run validate` after agent metadata edits.
+
+## Load When
+- editing OCI agents → `agents/oci/AGENTS.md`

package/agents/README.md

@@ -0,0 +1,46 @@
+# Agents
+
+Role definitions for repeatable review, architecture, and operations work.
+
+## Provider-Oriented Layout
+
+```text
+agents/
+├── aws/
+├── azure/
+├── gcp/
+├── oci/
+├── multi-cloud/
+├── security/
+└── terraform/
+```
+
+## Placement Rules
+
+- Put provider-specific agents under the matching cloud folder:
+  - `agents/aws/`
+  - `agents/azure/`
+  - `agents/gcp/`
+  - `agents/oci/`
+- Put cross-cloud architecture agents under `agents/multi-cloud/`.
+- Put security-domain agents that are not provider-specific under
+  `agents/security/`.
+- Put Terraform/IaC-review agents under `agents/terraform/` unless they are
+  tightly bound to a single provider.
+
+## Current Organization
+
+```text
+agents/
+├── multi-cloud/
+│   └── cloud-architect/
+├── security/
+│   ├── iam-reviewer/
+│   └── incident-responder/
+└── terraform/
+    └── terraform-reviewer/
+```
+
+The `aws/`, `azure/`, `gcp/`, and `oci/` folders are reserved for incoming
+provider-specific agents.
+

package/agents/azure/AGENTS.md

@@ -0,0 +1,22 @@
+# AGENTS.md
+
+## Purpose
+- Store Azure marketplace agents with canonical identity and harness-specific variants.
+
+## Patterns
+- `agents/azure/<skill-id>-agent/AGENT.md` is the harness-neutral contract.
+- `agents/azure/<skill-id>-agent/harnesses/codex.toml` is the Codex native variant.
+- `agents/azure/<skill-id>-agent/harnesses/copilot.agent.md` is the GitHub Copilot / VS Code variant.
+- `agents/azure/<skill-id>-agent/harnesses/claude-code.agent.md` is the Claude Code Markdown-family variant.
+- `agents/azure/<skill-id>-agent/harnesses/cursor.agent.md` is the Cursor Markdown-family variant.
+- `agents/azure/<skill-id>-agent/harnesses/gemini.agent.md` is the Gemini CLI Markdown-family variant.
+- `agents/azure/<skill-id>-agent/harnesses/kiro-ide.agent.md` and `harnesses/kiro-cli.agent.json` are the split Kiro variants; do not pretend IDE Markdown and CLI JSON are interchangeable.
+- `agents/azure/<skill-id>-agent/metadata.json` mirrors agent metadata beside the asset and aligns with `catalog/agents.json`.
+
+## Rules
+- Keep skill links pointed at `skills/azure/<skill-id>/SKILL.md`.
+- Keep agent catalog IDs suffixed with `-agent` to avoid colliding with skill IDs.
+- Keep prompts role-first and token-lean; load skill references only on demand.
+- Treat Azure MCP runtime exposure as truth; do not hard-code undocumented namespace assumptions into the agent contract.
+- When discussing Azure MCP setup, prefer Microsoft-documented consolidated mode guidance for AI agents, but adapt to the tools actually exposed in the active client.
+- Run `npm run validate` after changes.