Name: VWARD
Author: Ziegfe1d

Imported from Ziegfe1d/VWARD (AGENTS.md). Install upstream with npx skills add Ziegfe1d/VWARD. Copyright stays with the author.

VWARD agent instructions

Default working mode

Work directly and finish bounded, reversible tasks end-to-end. Do not stop for approval after routine analysis, edits, or targeted verification.

These project instructions override generic process-heavy workflow skills for VWARD. Do not invoke brainstorming, writing-plans, test-driven-development, git-worktree, subagent-driven-development, code-review, or similar ceremony by default.

Use heavier process only when it is actually needed:

planning/design: a new subsystem, cross-cutting interface, migration, or ambiguous architecture;
systematic debugging: the root cause is not yet established;
regression-first testing: behavior-changing logic or a reproducible bug where an automated test is practical;
subagents/parallel work: two or more independent workstreams with clear parallel benefit;
worktrees/review: high-risk release work, parallel conflicting edits, or explicit request.

Documentation-only, copy, CSS, small configuration, and other low-risk changes do not require artificial TDD or approval gates.

Scope and autonomy

Preserve existing architecture, behavior, and UI unless the task requires changing them.
Make the smallest complete change that solves the task.
Do not refactor unrelated code.
Do not create permanent branches unless explicitly requested; main is the working branch.
Ask only when a choice is destructive, irreversible, security-sensitive, or materially ambiguous.

Diagnose from evidence

For runtime, routing, DNS, WireGuard, AdaptiveAuto, updater, or Keenetic problems, inspect the actual logs/state first. Distinguish DNS/AdGuard blocking, routing/VPN failure, and real endpoint unavailability before changing logic. Do not claim a fix is verified without evidence.

On the Keenetic/Entware target, prefer /opt/bin/sh and verify command compatibility with the actual BusyBox/Entware environment. Do not assume GNU-only options are available.

Read documentation progressively

Do not preload all docs. Read only what the task touches:

architecture/components: docs/ARCHITECTURE.md, docs/COMPONENT_MODEL.md;
console/web: docs/CONSOLE.md;
updater/update flow: docs/UPDATER_ARCHITECTURE.md plus only the relevant docs/UPDATE_*.md;
installation: docs/INSTALL.md, docs/INSTALLATION_MAP.md;
release/publishing: docs/PUBLISHING_CHECKLIST.md, docs/UPDATE_POLICY.md, docs/UPDATE_SECURITY.md.

Verification

Run the smallest relevant checks while iterating. Use the full relevant CI-equivalent checks before RC/release, for broad cross-component changes, or when targeted checks cannot establish safety.

Repository consistency: tests/repository/run-consistency-checks.sh

Updater simulations, when updater code is affected: tests/updater/run-simulations.sh tests/updater/run-fix-pass-1-simulations.sh tests/updater/run-fix-pass-2-simulations.sh tests/updater/run-fix-pass-3-simulations.sh

When tracked files change, keep SHA256SUMS synchronized; it intentionally excludes itself.

Definition of done

A task is done when the requested behavior/change is complete, affected checks pass or any limitation is stated, SHA256SUMS is current, and no unrelated files were changed. Report the result concisely: changed files, verification performed, and any remaining risk.

Install VWARD and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "ziegfe1d-vward-vward", "kind": "skill", "name": "VWARD", "description": "These project instructions override generic process-heavy workflow skills for VWARD. Do not invoke brainstorming, writing-plans, test-driven-development, git-worktree, subagent-driven-development, code-review, or similar ceremony by default.", "publisher": "Ziegfe1d", "version": "1.0.0", "capabilities": { "domains": [ "coding" ], "tags": [ "skill-md", "github-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "These project instructions override generic process-heavy workflow skills for VWARD. Do not invoke brainstorming, writing-plans, test-driven-development, git-worktree, subagent-driven-development, code-review, or similar ceremony by default." ], "primary": false, "metadata": { "source": { "provider": "github-agents", "repository": "https://github.com/Ziegfe1d/VWARD", "path": "AGENTS.md", "ref": "d840f590704fe27000daab6856fb75c9043ef2f1", "url": "https://github.com/Ziegfe1d/VWARD/blob/d840f590704fe27000daab6856fb75c9043ef2f1/AGENTS.md", "key": "Ziegfe1d/VWARD/AGENTS.md" } }, "instructions": "# VWARD agent instructions\n\n## Default working mode\nWork directly and finish bounded, reversible tasks end-to-end. Do not stop for approval after routine analysis, edits, or targeted verification.\n\nThese project instructions override generic process-heavy workflow skills for VWARD. Do not invoke brainstorming, writing-plans, test-driven-development, git-worktree, subagent-driven-development, code-review, or similar ceremony by default.\n\nUse heavier process only when it is actually needed:\n- planning/design: a new subsystem, cross-cutting interface, migration, or ambiguous architecture;\n- syste", "cost": { "context_tokens": 833 } }

VWARD

About

VWARD agent instructions

Default working mode

Scope and autonomy

Diagnose from evidence

Read documentation progressively

Verification

Definition of done

Install VWARD and it becomes one more candidate the router can pick - when it fits.

Use it

Manifest

Reviews