docs(agents): introduce ARCHITECTURE.md + lessons + bash/powershell rules; misc lint and CI fixes

[szymonos]

Summary

• Agent harness foundation. AGENTS.md slimmed to a compound-knowledge index pointing at new files: ARCHITECTURE.md (host/guest split, distro detection, hook inventory, test layout, recipes), design/lessons.md (4 postmortems mined from fix(*) commits — POSIX guard, SSH probe, safe.directory, BSD regex), .claude/rules/bash-style.md and .claude/rules/powershell-style.md (auto-load on matching globs). Adds the per-repo /second-opinion skill with a linux-setup-scripts-targeted REVIEW-BRIEF.md.
• fix(sh): install_uv — switch from the brittle --system-certs flag to UV_SYSTEM_CERTS=true UV_NATIVE_TLS=true env vars on uv self update; also quote "$HOME/.local/bin/uv".
• chore(ci): repo_checks — runner from ubuntu-latest to ubuntu-slim.
• chore(lint): PREK_RUN macro — centralizes the stage-all → prek → restore-paths flow shared by make lint / lint-diff / lint-all, so auto-fixes land in the working tree without losing the prior staging (path-level; partially-staged hunks become fully staged). Extracts global shellcheck excludes (SC1090/SC2139/SC2148/SC2155/SC2174) into .shellcheckrc so VS Code, CLI, and pre-commit all see the same config; bumps the pre-commit hook to the shellcheck-py mirror.

Test plan

• make lint-diff — all 13 pre-commit hooks green on the consolidated commits
• /second-opinion (gpt-5.3-codex) reviewed the WIP — 1 confirmed bug (broken .claude/CLAUDE.md symlink target) fixed before consolidation, 1 warning dismissed as pre-existing
• Copilot PR review (10 threads) — 6 fixed (lessons.md commit links demoted to bare hashes, .shellcheckrc rationale inlined, install_uv path quoted, Makefile comment scoped to "paths"), 4 dismissed (mktemp false positive, ubuntu-slim valid in env, partial-staging pre-existing)
• Reviewer: verify .claude/CLAUDE.md resolves to root AGENTS.md (cat .claude/CLAUDE.md should show the AGENTS.md content)
• Reviewer: skim ARCHITECTURE.md § 5 recipes — are they accurate descriptions of how to extend this repo?
• Reviewer: skim design/lessons.md entries — each maps to a real commit and a real durable rule

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR adds a more structured “agent harness” documentation set (architecture + lessons learned + per-language style rules), introduces a /second-opinion Claude skill backed by GitHub Copilot CLI, and makes several lint/CI reliability tweaks (Makefile lint macro, ShellCheck config centralization, uv update behavior, and a workflow runner change).

Changes:

• Introduces new docs: ARCHITECTURE.md (repo layout/recipes) and design/lessons.md (incident → rule postmortems), and refactors AGENTS.md into an index.
• Adds .claude rules and a /second-opinion skill with parsing/brief tooling (review_brief.py, REVIEW-BRIEF.md).
• Updates lint/CI plumbing: PREK_RUN Makefile macro, moves ShellCheck excludes to .shellcheckrc, swaps ShellCheck pre-commit repo, tweaks uv self-update invocation, and changes the GitHub Actions runner label.

Reviewed changes

Copilot reviewed 14 out of 15 changed files in this pull request and generated 10 comments.

Show a summary per file

File	Description
Makefile	Adds PREK_RUN macro to preserve staged paths while running prek and reuses it across lint targets.
design/lessons.md	New postmortem-style lessons doc with commit references and durable rules.
ARCHITECTURE.md	New repo architecture and extension “recipes” documentation.
AGENTS.md	Slims to a compound-knowledge index pointing to the new docs and style rules.
.shellcheckrc	Centralizes global ShellCheck disables for consistent CLI/IDE/pre-commit behavior.
.pre-commit-config.yaml	Switches ShellCheck hook to the shellcheck-py mirror and relies on .shellcheckrc for excludes.
.gitignore	Ignores local Claude/OpenWolf-related files and settings.
.github/workflows/repo_checks.yml	Changes CI runner label for pre-commit checks workflow.
.claude/skills/second-opinion/SKILL.md	Adds skill definition + workflow for invoking GitHub Copilot CLI review and parsing results.
.claude/skills/second-opinion/scripts/review_brief.py	Adds tooling to check/discover/parse the second-opinion review brief.
.claude/skills/second-opinion/REVIEW-BRIEF.md	Adds repo-targeted review brief and output format for heterogeneous-model review.
.claude/rules/powershell-style.md	Adds PowerShell formatting/naming/error-handling conventions for agent auto-loading.
.claude/rules/bash-style.md	Adds Bash portability/style conventions for agent auto-loading.
.assets/provision/install_uv.sh	Adjusts uv self update to use env vars instead of the removed/brittle --system-certs flag.

[Copilot]

Pull request overview

Copilot reviewed 14 out of 15 changed files in this pull request and generated 3 comments.