feat: add agentic-architect plugin with architecture design and plugin review skills

* feat: add agentv-plugin-review skill and eval

Add a new skill to agentv-dev for reviewing AI plugin PRs across three
dimensions: skill quality, eval correctness, and workflow architecture.

Includes:
- SKILL.md with 5-step review process
- references/eval-checklist.md for eval file review
- references/workflow-checklist.md for OpenSpec-based workflow review
- 8-test eval covering missing evals, naming, assertions, file paths,
  repeated inputs, hard gates, factual contradictions, and command refs

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: add workspace-based eval with mock plugin and planted issues

Rewrite eval to use pi-cli target with a mock workspace containing a
deploy-auto plugin with planted issues: missing eval, wrong extensions,
no assertions, relative paths, repeated inputs, missing hard gates,
factual contradiction, nonexistent command reference, and hardcoded paths.

Uses shorthand input, skill-trigger assertions, and workspace template
instead of loading SKILL.md directly.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: use default target instead of invalid pi-cli

The pi-coding-agent provider is configured as the 'default' target
in .agentv/targets.yaml. pi-cli is not a valid target name.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: add pi-cli target and use it for plugin-review eval

Add pi-cli as a named target in .agentv/targets.yaml using the
pi-coding-agent provider. Update the plugin-review eval to use it.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: add skill quality checklist from Superpowers writing-skills

Add references/skill-quality-checklist.md covering CSO (Claude Search
Optimization), description anti-patterns, content quality, file org,
flowchart usage, and discipline-enforcing skill checks.

Update SKILL.md Step 2 to reference the checklist and integrate key
CSO principles (description must not summarize workflow, etc).

Remove plugin.json from review table (handled by CI).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: create agentic-architect plugin with architecture design and plugin review skills

Move agent-plugin-review from agentv-dev to new agentic-architect plugin.
Add agent-architecture-design skill with species taxonomy (Coding Harness,
Dark Factory, Auto Research, Orchestration) and workflow patterns from
OpenSpec, Superpowers, and Compound Engineering.

New plugin structure:
- agent-architecture-design: diagnose problem, select species, design workflow
- agent-plugin-review: review skills, evals, and workflow architecture

Register in both .claude-plugin and .github marketplace.json.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: add pi-coding-agent target and fix pi-cli provider

- pi-cli target now uses provider: pi-cli (matching its name)
- pi-coding-agent target added as explicit alias matching default config

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: remove codex cwd and log_dir from targets.yaml

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* refactor: replace "species" terminology with industry-standard "agentic design patterns"

Rename species-taxonomy.md to agentic-design-patterns.md. Replace all
"Species A/B/C/D" labels with standard terms: Single-Agent Iterative Loop,
Autonomous Pipeline, Optimization Loop, Multi-Agent System.

Based on Andrew Ng's agentic design patterns framework and industry
conventions (cognitive architectures, agentic workflows).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: remove name attribution from design patterns reference

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: rename evals/architecture to evals/self

These evals validate the agentv repo itself, not architecture patterns.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: add deterministic linting scripts for eval and plugin review

- agentv-eval-review: new skill in agentv-dev with lint_eval.py script
  for .eval.yaml naming, file paths, assertions, prose detection, repeated inputs
- agent-plugin-review: add lint_plugin.py script for frontmatter, hardcoded
  paths, version printing, missing evals, referenced files, command refs
- Remove eval-checklist.md from agent-plugin-review (handled by agentv-eval-review)
- Both SKILLs are now minimal wrappers: run script first, then LLM judgment

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: add self-consistency and cross-file consistency checks

Address remaining LLM-judgment gaps from PR #324 review:
- Internal self-consistency (skill contradicting itself)
- Cross-file consistency (filenames/tools matching across skills and evals)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: simplify consistency check to single line

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: remove .entire, .claude, .github/hooks, .opencode/plugins

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: rename evals/self/dataset.eval.yaml to eval.yaml

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: remove execution target from self eval

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: flag manual routing workarounds as weak description signal

AGENTS.md or instruction files with heavy TRIGGER/ACTION routing tables
indicate skill descriptions aren't enabling auto-discovery. Good
descriptions make manual routing unnecessary.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: add external dependencies check to plugin review

Skills that depend on MCP servers, external repos, specific directory
layouts, or services should declare these explicitly.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* refactor: remove external dependencies check, add agentskills.io references

External dependencies is not a best practice per agentskills.io — the
compatibility field and gotchas sections handle this. The real issue
(hardcoded paths) is already caught by lint_plugin.py.

Add agentskills.io specification, best practices, and description
optimization as external references for discoverable guidance.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: add agentskills.io using-scripts reference

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* ci: add markdown link validation workflow

Check relative markdown links on push to main and PRs using lychee.
Offline mode only (local links, not HTTP URLs).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: repair 12 broken markdown links across 3 files

- apps/cli/README.md: add ../../ prefix for repo-root references
- examples/showcase/offline-grader-benchmark/README.md: fix CLAUDE.md depth
- packages/eval/README.md: agentv-eval-builder → agentv-eval-writer

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: exclude apps/examples symlink from link checker

apps/examples is a symlink to ../examples which causes lychee to
resolve relative paths incorrectly (../../../CLAUDE.md resolves to
apps/CLAUDE.md instead of repo root CLAUDE.md).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: remove apps/examples symlink

Unused symlink to ../examples that confused lychee link checker.
examples/ is accessible directly at repo root.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* refactor: symlink apps/cli/README.md to root README

apps/cli/README.md is published to npm and was a stale copy of the
root README with diverging relative links. Replace with a symlink so
it stays in sync automatically.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: christso

.agentv/targets.yaml

@@ -10,11 +10,21 @@ targets:
     system_prompt: "Answer directly based on the information provided."
     grader_target: gemini-flash
 
+  - name: pi-cli
+    provider: pi-cli
+    grader_target: gemini-flash
+
+  - name: pi-coding-agent
+    provider: pi-coding-agent
+    subprovider: openrouter
+    model: z-ai/glm-4.7
+    api_key: ${{ OPENROUTER_API_KEY }}
+    system_prompt: "Answer directly based on the information provided."
+    grader_target: gemini-flash
+
   - name: codex
     provider: codex
     grader_target: gemini-llm
-    cwd: ${{ CODEX_WORKSPACE_DIR }}            # Where scratch workspaces are created
-    log_dir: ${{ CODEX_LOG_DIR }}              # Optional: where Codex CLI stream logs are stored (defaults to ./.agentv/logs/codex)
     log_format: json                    # Optional: 'summary' (default) or 'json' for raw event logs
 
   - name: gemini-llm

.claude-plugin/marketplace.json

@@ -16,6 +16,11 @@
       "name": "agentv-claude-trace",
       "description": "Session tracing plugin — exports Claude Code session traces via OpenTelemetry",
       "source": "./plugins/agentv-claude-trace"
+    },
+    {
+      "name": "agentic-architect",
+      "description": "Design and review AI agent architectures — species selection, workflow patterns, and plugin quality review",
+      "source": "./plugins/agentic-architect"
     }
   ]
 }

.claude/settings.json

@@ -16,6 +16,11 @@
       "name": "agentv-claude-trace",
       "description": "Session tracing plugin — exports Claude Code session traces via OpenTelemetry",
       "source": "./plugins/agentv-claude-trace"
+    },
+    {
+      "name": "agentic-architect",
+      "description": "Design and review AI agent architectures — species selection, workflow patterns, and plugin quality review",
+      "source": "./plugins/agentic-architect"
     }
   ]
 }

.entire/.gitignore

@@ -0,0 +1,23 @@
+name: Validate
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  links:
+    name: Check Links
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Check relative markdown links
+        uses: lycheeverse/lychee-action@v2
+        with:
+          args: >-
+            --offline
+            --no-progress
+            --glob-ignore-case
+            --root-dir .
+            "**/*.md"