Add code-review skill for cross-agent diff review

[blarghmatey]

What are the relevant tickets?

N/A — tracked as an agent-kit workflow project task, no GitHub issue.

Description (What does it do?)

Adds skills/process/code-review/SKILL.md, giving pi/copilot/opencode
sessions the same diff-review capability Claude Code gets from its
built-in /code-review — correctness bugs plus
reuse/simplification/efficiency cleanups over a diff.

• Ports the four-dimension rubric (correctness, simplification,
  efficiency, reuse) and the "verify before reporting" discipline
  already established in address-pr-feedback (address-pr-feedback: verify claims, prove fixes, slow down on human review #251) — a finding
  gets re-checked against the actual code before it ships.
• Target resolution covers no-arg (working tree diff), branch, path,
  and PR-number inputs, stated as prose rather than CLI flags, since
  not every platform this skill runs on has an argument-passing
  mechanism.
• No effort-tier flag either, for the same reason — depth is read from
  the request ("be thorough" widens the pass) rather than a parameter.
• Report-only by default; fixes only apply after the review is
  reported, and only when the request already asked for a fix.
• No MCP or witan dependency — needs only git (and gh for a
  PR-number target), so it installs and runs the same on every
  platform in the catalog.
• references/dimensions.md and references/findings-format.md
  carry the worked examples, mirroring address-pr-feedback's
  SKILL.md + references anatomy.
• Registered in agent-config.toml's [skills] table and
  [profiles.process], and added to both skills/README.md and
  skills/process/README.md's catalogs, per AGENTS.md's "Adding a
  Skill" checklist.

How can this be tested?

• prek run against all changed files (markdownlint, TOML checks,
  trailing-whitespace, secret detection) — all passed.
• agent-kit validate agent-config.toml — no missing/mismatched
  entries; the only output is expected drift for skills not yet
  applied to this machine's install directories.
• Manually walked the skill's own steps in this Claude Code session
  against a real diff (43fc1c6, path-scoped to
  mcp/servers/witan/witan/cli/{local_dispatch.py,_common.py} and
  config.py): the path-scoped target-resolution rule worked
  mechanically, and I produced an actual findings-table pass rather
  than only the worked examples in references/. Outcome: no
  high-confidence findings survived — one plausible-looking
  correctness concern (a non-remote target matching by match_paths
  could mark local-store use "deliberate" and suppress the refusal
  even when a differently-matching target elsewhere declares the real
  deployment) turned out to hinge on match_target()'s precedence
  logic in witan_core.target_config, outside the 3-file scope of
  that diff — so per the skill's own verification-pass rule, it didn't
  ship as a finding. That's the report-nothing-if-nothing-survives
  path working as designed, not a skipped step.
• Not done: a live second-platform run (pi/copilot/opencode), called
  for in the task's definition of done. Flagging this as open rather
  than claiming it — reviewer should treat that verification as
  outstanding.

Additional Context

No scripts/ directory — steps need only git diff, a rubric, and a
markdown table, matching the "no MCP, no scripts" premise the spec
called for. If a reviewer finds the skill needs one in practice,
that's a signal the premise needs revisiting, not something to add
silently.

[Copilot]

Pull request overview

Adds a portable cross-agent code-review skill with scope resolution, review dimensions, verification guidance, and standardized reporting.

Changes:

• Defines the code-review workflow and four-dimension rubric.
• Adds findings-format and review-dimension references.
• Registers and catalogs the new skill.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
skills/README.md	Adds the skill to the main catalog.
skills/process/README.md	Adds the skill to the process catalog.
skills/process/code-review/SKILL.md	Defines review scope, verification, reporting, and fix mode.
skills/process/code-review/references/findings-format.md	Documents the findings schema and examples.
skills/process/code-review/references/dimensions.md	Documents the four review dimensions.
agent-config.toml	Registers the skill in the process profile.

Suppressed comments (2)

skills/process/code-review/SKILL.md:42

• refs/remotes/origin/HEAD is optional and is often absent in local-only repositories or clones where the remote HEAD was not initialized. In those checkouts this command fails, so branch review cannot reach merge-base, despite the stated portability. Define a fallback (for example, discover another remote's HEAD or ask the user for the base branch) rather than assuming this ref exists.

2. **A branch name** — diff against where it forked from the default
   branch, not a plain two-dot diff: detect the default branch (`git
   symbolic-ref refs/remotes/origin/HEAD`), find the merge base (`git
   merge-base <default> <branch>`), then `git diff <merge-base>...<branch>`.

skills/process/code-review/SKILL.md:44

• A path-only review uses git diff -- <path>, which shows only unstaged changes. If the requested path is staged, the skill reports an empty diff and misses the change; use HEAD for a working-tree path target and account for untracked files too.

3. **A path** — `git diff -- <path>` (or the branch-scoped equivalent above
   if a branch was also named), same default-branch detection.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[blarghmatey]

Addressed all review feedback in 8d61e36: 3 threads fixed and resolved (untracked-file gap, verification-pass/dimensions.md conflict, depth/verification drop-rule conflict). Copilot's 2 suppressed comments (missing origin/HEAD fallback, path-diff missing staged changes) had no separate thread to resolve but are fixed in the same commit.