SEO rework: meta description, comparison table, keyword-rich title

.agents/skills/feature-exploration/SKILL.md

@@ -0,0 +1,35 @@
+---
+name: feature-exploration
+description: Explore unfamiliar APIs, libraries, or implementation behavior with minimal reproducible examples and documented findings. Use when implementation details are unclear and can be resolved by reading docs, inspecting code, and running focused experiments.
+---
+
+# Feature Exploration
+
+Use this developer skill when a task depends on API behavior that is unclear from memory or partially documented. The goal is to resolve uncertainty before changing production code.
+
+## Workflow
+
+1. State the uncertainty: name the API, behavior, or integration detail that needs proof.
+2. Read the closest authoritative docs or source code before experimenting.
+3. Build the smallest reproducible example that answers the question.
+4. Run it in the project environment when it imports project code or dependencies.
+5. Iterate only until the behavior is understood.
+6. Record the finding in the final answer or in the durable project docs if it affects future work.
+7. Apply the production change using the confirmed behavior.
+
+## Guardrails
+
+- Do not create throwaway files in tracked locations. If a temporary note is needed, use `.scratch/`.
+- Do not leave exploratory scripts behind unless they become real tests or documented examples.
+- Prefer adding a proper test under `causalpy/tests/` when the discovered behavior is important to preserve.
+- Keep experiments narrow; avoid broad refactors while investigating.
+- Follow the repository environment rules in `AGENTS.md` for commands that import CausalPy, PyMC, PyTensor, matplotlib, or repo tooling.
+
+## Output
+
+When using this skill, report:
+
+- The uncertainty investigated.
+- The evidence gathered.
+- The conclusion.
+- Any production change or test that now relies on that conclusion.

.github/skills/github-issues/reference/bug-report.md → .agents/skills/github-issues/reference/bug-report.md

@@ -32,6 +32,7 @@ Bug-specific template:
 <Clear, concise description of the bug>
 
 ## Steps to Reproduce
+
 1. <Step 1>
 2. <Step 2>
 
@@ -42,6 +43,7 @@ Bug-specific template:
 <What actually happens>
 
 ## Environment
+
 - CausalPy version:
 - Python version:
 - OS:

.github/skills/github-issues/reference/issue-evaluation.md → .agents/skills/github-issues/reference/issue-evaluation.md

@@ -65,6 +65,7 @@ Create `.scratch/issue_comments/issue-<number>-evaluation.md` with:
 <Restate issue + status>
 
 ### Current Relevance
+
 - Code references checked:
 - Related PRs:
 - Changes since issue creation:
@@ -73,8 +74,10 @@ Create `.scratch/issue_comments/issue-<number>-evaluation.md` with:
 <Key points from comments>
 
 ### Recommendation
+
 - **Status:** <Ready / Needs info / Blocked / Can be closed>
 - **Proposed next steps:** <Actionable items>
+
 ```
 
 Present the draft for user approval before posting. Do not hard-wrap lines in markdown drafts; keep paragraphs on a single line.

.github/skills/github-issues/reference/parent-child-issues.md → .agents/skills/github-issues/reference/parent-child-issues.md

@@ -128,6 +128,7 @@ gh api graphql -f query='query {
 Do **not** apply fallback automatically.
 
 When native sub-issues are unavailable, first inform the user:
+
 - what capability check failed (missing `parent`/`subIssues` fields or no `addSubIssue` mutation)
 - that native hierarchy cannot be created in this repo/account context
 - that a markdown tracking-issue fallback is available
@@ -137,8 +138,10 @@ Then explicitly ask whether they want to proceed with fallback. Only continue af
 Use a tracking issue body with markdown task links:
 
 ```md
+
 - [ ] #123 Child issue one
 - [ ] #124 Child issue two
+
 ```
 
 This renders correctly on GitHub and preserves parent-child tracking even without native hierarchy.

.github/skills/pr-to-green/SKILL.md → .agents/skills/pr-to-green/SKILL.md

@@ -81,6 +81,7 @@ Use subagents (via the Task tool) when work is noisy, broad, or high-risk. Each
 Trigger when CI logs are long/noisy, failures span multiple jobs, or failure family is unclear.
 
 Handoff (include in Task prompt):
+
 - PR number and branch name
 - Failed job names and relevant log snippets or `gh run view` output
 - Ask for: root cause ranking, fix-first order, and minimal patch plan
@@ -90,6 +91,7 @@ Handoff (include in Task prompt):
 Trigger when rebase/merge produces non-trivial conflicts, intent differs across branches, or conflict count is high.
 
 Handoff (include in Task prompt):
+
 - Target branch and merge/rebase direction
 - Full conflict list (`git diff --name-only --diff-filter=U`) and key conflict markers/snippets
 - Ask for: conflict risk map, lowest-risk resolution order, and explicit escalations
@@ -99,11 +101,13 @@ Handoff (include in Task prompt):
 Do not auto-resolve and continue silently when conflicts are highly complex. Instead, stop and request maintainer input with a brief report.
 
 Escalate by default when:
+
 - `.ipynb` conflicts are messy (overlapping cell content plus metadata/output churn)
 - conflict resolution would drop one branch's meaningful behavior
 - conflicts touch core contracts and intent is ambiguous (experiments/models/tests/docs all changed around same behavior)
 
 Escalation report must include:
+
 - conflicted file list and risk class
 - resolution options (1-2) with trade-offs
 - recommended next step and what decision is needed from maintainer
@@ -113,12 +117,14 @@ Escalation report must include:
 Do not keep patching silently when greening a PR turns into broader compatibility work outside the PR's feature surface. Stop after root-cause identification and ask the maintainer whether to continue in the PR or split the work into a separate branch/PR from `main`.
 
 Escalate by default when:
+
 - the failing checks are caused by third-party API or version drift in shared/core code
 - the fix would benefit multiple open PRs or the default branch, not just the current PR
 - the next fix would modify shared integrations beyond the feature the PR is introducing
 - successive CI reruns keep exposing new failure families outside the original PR scope
 
 Scope-drift report must include:
+
 - the original PR goal and the newly discovered broader issue
 - which files are feature-specific versus shared/core compatibility files
 - options: patch in the PR, split a separate compatibility PR, or pause for maintainer direction

.github/skills/pr-workflows/reference/pre-commit.md → .agents/skills/pr-workflows/reference/pre-commit.md

@@ -13,6 +13,9 @@ prek run --files path/to/file.py
 
 # Full verification before push/commit
 prek run --all-files
+
+# Python source/test changes: local patch coverage gate
+make test-patch-cov
 ```
 
 ## If hooks modify files

.github/skills/review-pr/SKILL.md → .agents/skills/review-pr/SKILL.md

@@ -65,7 +65,7 @@ Read these when the PR touches the relevant surface:
 
 ## CausalPy Review Norms
 
-- Before reviewing code, read `AGENTS.md` and relevant local context. For docs-heavy PRs, also inspect `docs/source/notebooks/index.md`; for process-sensitive PRs, inspect `CONTRIBUTING.md` when present.
+- Before reviewing code, read `AGENTS.md` (workflow) and `ARCHITECTURE.md` (design) for core changes. For docs-heavy PRs, also inspect `docs/source/notebooks/index.md`; for process-sensitive PRs, inspect `CONTRIBUTING.md` when present.
 - Use `$CONDA_EXE run -n CausalPy <command>` for commands that import project code, run tests, build docs, or invoke repo tooling. `AGENTS.md` defines how to detect or set `CONDA_EXE`; if it is unset, inspect that environment guidance before running project commands.
 - Use full permissions for commands that import PyMC, PyTensor, or matplotlib to avoid false sandbox failures.
 - During review, prefer targeted local checks that match the changed surface. If you edit code or prepare a commit, run `prek run` during iteration and `prek run --all-files` before handoff unless the user explicitly says not to.
@@ -79,6 +79,7 @@ Use this structure:
 
 ```markdown
 ## Findings
+
 - [severity] `path`: issue, why it matters, and what should change.
 
 ## Merge Readiness

.github/skills/review-pr/resources/maintenance.md → .agents/skills/review-pr/resources/maintenance.md

@@ -42,6 +42,7 @@ Each recurring pattern should include:
 
 - Canonical example: <PR reference and concrete instance>.
 - How to spot: <specific diagnostic prompt>.
+
 ```
 
 ## When to Prune

.github/skills/review-pr/resources/review-comments.md → .agents/skills/review-pr/resources/review-comments.md

@@ -57,6 +57,7 @@ Thanks for the PR. I have comments grouped by severity below.
 
 - [Specific positive about the implementation, tests, docs, or scope.]
 - [Another specific positive when available.]
+
 ```
 
 Keep positives specific. Generic praise adds little, but concrete reinforcement helps contributors preserve the good parts while fixing issues.

.cursor/agents/ci-failure-investigator.md

@@ -25,12 +25,14 @@ When invoked:
 5. Recommend exact verification commands to confirm fixes.
 
 CausalPy-specific requirements:
+
 - Respect AGENTS.md conventions: CausalPy env via `$CONDA_EXE run -n CausalPy <command>`.
 - For commands importing PyMC/PyTensor/matplotlib or running pytest/doctest, explicitly note full permission requirements.
 - Prefer targeted reruns before full suite reruns.
 - If behavior changed, require proper pytest updates under `causalpy/tests/` (no throwaway scripts).
 
 Output format (strict):
+
 - Failing checks summary
 - Root-cause ranking (highest leverage first)
 - Minimal patch plan

.cursor/agents/merge-conflict-analyst.md

@@ -15,6 +15,7 @@ Operating mode:
 - Escalate ambiguous or high-risk conflicts to maintainer decision.
 
 When invoked:
+
 1. Inventory conflicted files and classify each as:
    - mechanical (format/import ordering/adjacent edits)
    - semantic (behavior/API/test/docs meaning may differ)
@@ -28,11 +29,13 @@ When invoked:
 4. After resolution, recommend focused verification commands and required tests.
 
 Hard escalation rules (must not auto-resolve):
+
 - `.ipynb` conflicts with many overlapping cell/output/metadata changes
 - conflicts affecting experiment/model contracts (`causalpy/experiments/`, `causalpy/pymc_models.py`, `causalpy/skl_models.py`) where intent is unclear
 - conflicts that require dropping one side's behavior without explicit maintainer approval
 
 Notebook-specific handling:
+
 - Prefer preserving semantic cell content and minimizing output/metadata churn.
 - If notebook conflict is messy, produce a maintainer report:
   - conflicting notebook paths
@@ -41,11 +44,13 @@ Notebook-specific handling:
   - recommended manual next action
 
 CausalPy requirements:
+
 - Respect AGENTS.md conventions and avoid destructive git operations.
 - Never discard unknown user changes.
 - If behavior changes, require pytest updates in `causalpy/tests/`.
 
 Output format (strict):
+
 - Conflict map (file -> risk class)
 - Proposed resolutions (ordered low-risk to high-risk)
 - Escalations requiring maintainer input

.github/workflows/ci.yml

@@ -26,12 +26,12 @@ jobs:
           python-version: "3.12"
       - name: Run prek on changed files in PRs
         if: github.event_name == 'pull_request'
-        uses: j178/prek-action@6ad80277337ad479fe43bd70701c3f7f8aa74db3 # v2.0.3
+        uses: j178/prek-action@bdca6f102f98e2b4c7029491a53dfd366469e33d # v2.0.4
         with:
           extra-args: --from-ref ${{ github.event.pull_request.base.sha }} --to-ref ${{ github.event.pull_request.head.sha }}
       - name: Run prek on all files on pushes
         if: github.event_name != 'pull_request'
-        uses: j178/prek-action@6ad80277337ad479fe43bd70701c3f7f8aa74db3 # v2.0.3
+        uses: j178/prek-action@bdca6f102f98e2b4c7029491a53dfd366469e33d # v2.0.4
         with:
           extra-args: --all-files
 
@@ -102,7 +102,7 @@ jobs:
           python ./docs/source/.codespell/notebook_to_markdown.py --tempdir tmp_markdown
           codespell
       - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@57e3a136b779b570ffcdbf80b3bdc90e7fab3de2 # v6.0.0
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 # v6.0.1
         with:
           token: ${{ secrets.CODECOV_TOKEN }} # use token for more robust uploads
           name: ${{ matrix.python-version }}