chore(skills): refactor process-issues to goal-oriented skill

.claude/commands/process-issues.md

@@ -1,71 +1,52 @@
-Process all open GitHub issues end-to-end. Do not stop until every issue is resolved or explicitly deferred.
-
-## Workflow
-
-For each open issue (`gh issue list --state open`), in order by issue number:
-
-### 0. Filter — owner/approval gate
-- Only process issues that meet ONE of:
-  - Created by `chaliy` (`gh issue view N --json author --jq '.author.login'`)
-  - Has a comment from `chaliy` approving it (`gh api repos/{owner}/{repo}/issues/N/comments --jq '.[].author.login' | grep -q chaliy`)
-- Skip all other issues silently — do not comment, do not close, just move to next
-
-### 1. Understand
-- Read the issue body and classify: bug, feat, test, chore, refactor, docs
-- Identify affected areas: parser, interpreter, builtins, vfs, network, git, python, tool, eval, security
-- Create a branch from main: `fix/issue-{N}-{short-slug}`
-
-### 2. Reproduce (bugs) / Scaffold (features)
-- Write a failing test first — spec test in `crates/bashkit/tests/spec_cases/` or unit test in the relevant module
-- Verify the test actually fails before proceeding
-
-### 3. Implement
-- Fix the bug or implement the feature
-- Keep changes minimal and focused
-
-### 4. Threat model review
-- If the change touches parser, interpreter, VFS, network, git, or user input:
-  - Check `specs/006-threat-model.md` for applicable threat IDs
-  - Add mitigations or update existing entries if needed
-  - Add security tests in `tests/threat_model_tests.rs` if new attack surface
-
-### 5. Exception & edge-case tests
-- Add negative tests: invalid input, empty input, boundary values, error paths
-- Add security tests if touching sandboxing or permissions (`specs/005-security-testing.md`)
-- Ensure both positive and negative scenarios covered
-
-### 6. Update specs
-- If behavior changes, update the relevant spec in `specs/`
-- Update `specs/009-implementation-status.md` if feature status changed
-
-### 7. Verify
-- Run `just pre-pr` (fmt + clippy + test + vet)
-- Fix any failures before proceeding
-
-### 8. Commit & PR
-- Commit with conventional format: `type(scope): description` — reference `Closes #N`
-- Push branch, create PR with summary + test plan
-- Wait for CI green
-
-### 9. Merge & close
-- Squash-merge the PR (`gh pr merge --squash --delete-branch`)
-- Add resolution comment on the issue if helpful
-- Rebase on latest main before starting next issue: `git checkout main && git pull origin main`
-- Move to next issue
-
-### 10. Review ignored tests
-After all issues are processed, scan for `#[ignore]` tests across the codebase:
-- `grep -rn '#\[ignore\]' crates/` to find all ignored tests
-- For each ignored test, determine why it was ignored (read surrounding comments, git blame)
-- Classify into: (a) can un-ignore now — underlying issue fixed, (b) blocked — still needs work, create/link issue, (c) intentionally ignored — e.g. slow, requires external resource
-- Un-ignore tests in category (a), run them, verify they pass
-- For category (b), ensure a tracking issue exists
-- Commit any un-ignored tests as `test: un-ignore {test_name}, now passing`
-- Create a single PR for all un-ignored tests (separate from issue PRs)
-
-## Rules
-- One issue = one PR. Do not bundle.
-- If an issue is unclear or not reproducible, comment asking for clarification and skip to next.
-- If a fix would be too large (>500 lines), split into sub-issues and link them.
-- Never skip the failing-test-first step for bugs.
-- Return to main between issues: `git checkout main && git pull origin main`
+Resolve all open GitHub issues. Each issue gets its own PR. Do not stop until every issue is resolved or deferred.
+
+## Arguments
+
+- `$ARGUMENTS` - Optional: specific issue number(s) to process. If omitted, process all open issues.
+
+## Goals
+
+Each goal below is an outcome to achieve per issue, not a script to follow. Use whatever tools and approaches make sense.
+
+### 1. Every actionable issue has a merged PR
+
+For each open issue, ordered by issue number, produce a **separate PR** that closes it.
+
+**CRITICAL: One issue = one PR. Never bundle multiple issues into a single PR. This is a hard requirement — no exceptions.**
+
+Only process issues authored by `chaliy` or approved by `chaliy` in comments. Skip all others silently.
+
+Key tools: `gh issue list --state open`, `gh issue view`
+
+### 2. Every bug fix proves the bug existed
+
+A failing test exists before the fix is applied. Verify the failure. Then fix. This is non-negotiable.
+
+For features, an acceptance test defines expected behavior before implementation.
+
+### 3. Every change is hardened
+
+Negative tests, edge cases, and security tests exist where applicable (parser, VFS, sandbox, permissions).
+
+Key references: `specs/006-threat-model.md`, `specs/005-security-testing.md`
+
+### 4. Specs and docs reflect reality
+
+If behavior changed, the relevant spec in `specs/` is updated. `specs/009-implementation-status.md` reflects current feature status.
+
+### 5. Every PR passes quality gates
+
+`just pre-pr` is green before creating the PR. CI is green before merging.
+
+### 6. Ignored tests are reviewed
+
+After all issues are processed, `#[ignore]` tests are scanned. Any that now pass are un-ignored in a separate PR.
+
+## Execution
+
+- Branch from latest main per issue
+- Conventional commits referencing `Closes #N`
+- Squash-merge each PR, return to main before starting the next issue
+- Unclear or non-reproducible issues → comment asking for clarification, skip to next
+- Too large (>500 lines) → split into sub-issues, link them
+- Use parallel agents for independent issues when possible