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>

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>

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>

…ugin 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>

- 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>

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

…ic 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>

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

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

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

- 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>

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>

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

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

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

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

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>

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>

…rences

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>

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

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>

- 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>

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>

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>

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>