Extract mechanical git/gh sequences from skill markdown into scripts or Make targets

[sebastientaggart]

Problem to Fix

The agent-facing skill files (skills/start.md, skills/submit-for-review.md, skills/deploy.md, and related files) embed multi-step git and gh plumbing inline — dirty-tree guards, fetch + hard-reset sequences, --body-file plumbing, label/milestone resolution tiers, follow-up-issue creation loops.

Asking the agent to execute each mechanical step in turn costs tokens, increases the surface area for mistakes (skipped guards, misquoted flags, forgotten temp files before gh ... --body-file), and forces every adapter (Claude/Cursor/Codex/Gemini) to re-derive the same shell sequence at runtime.

Why it Matters

• Tokens: each invocation re-loads the full skill markdown and re-narrates each step. A scripted equivalent is one tool call.
• Reliability: deterministic shell logic in a script behaves identically every time; agent-executed sequences can drift.
• Auditability: the dirty-tree + fetch + reset sequence added in Local integration branch (dev) accumulates SHA-divergent history vs origin/dev over time — make merge uses git pull where it needs git reset --hard #156 / PR Hard-reset integration branch after merge to prevent SHA divergence #158 is the prototype — easier to reason about as one shell target than as a chain of agent-issued commands.
• Adapter parity: a script is invoked identically across adapters and removes runtime re-derivation.

General Approach (tracer bullet)

This ticket is intentionally scoped to a single tracer-bullet extraction — the dirty-tree guard + git fetch + git reset --hard origin/<base> sequence currently inlined in /start (Case A Step 4 and Case B Step 4) and /deploy Step 1.

Concrete changes:

1. Move skills/* to skills/cc/* so per-skill-group scripts/ directories have a natural home.
2. Add skills/cc/scripts/sync-base-branch.py implementing the dirty-tree-guard + fetch + hard-reset sequence. Python over bash for arg handling, structured errors, and clear exit codes.
3. Wire the script into the three existing call-sites in skills/cc/start.md and skills/cc/deploy.md, replacing the inlined shell with a single invocation.

Everything else — extracting --body-file plumbing, follow-up-issue loops, label/milestone resolution, sync.py awareness of the new layout, broader skill-group support — is out of scope and tracked under #157.

Blocked on #157

This ticket is blocked on #157 (skill groupings). The skills/cc/ move requires sync.py to know where to find skills, which is #157's responsibility. Landing this PR before #157 would leave the project in a broken state. Resume work here once #157 has merged.

Complexity

Verification / QA effort: moderate

The change touches three skill call-sites across all four adapters and must be re-validated end-to-end (start → submit → deploy). Gate on ./sync.py --validate + a manual workflow run after #157's layout is in place.

Acceptance Criteria

• skills/ contents relocated under skills/cc/ (consistent with the layout Organize skills into opt-in groups so projects can curate which skill sets are active #157 establishes).
• skills/cc/scripts/sync-base-branch.py exists and implements the dirty-tree guard + git fetch + git reset --hard origin/<base> sequence with clear exit codes.
• /start (both Case A Step 4 and Case B Step 4) and /deploy Step 1 invoke the script as a single command instead of enumerating the individual shell steps.
• Behavior is identical end-to-end: dirty-tree guards still abort, hard-reset still runs against origin/<base>.
• ./sync.py --validate passes and all four adapter outputs continue to render correctly.

[sebastientaggart]

Scope tightened to a tracer-bullet extraction and deferred behind #157.

Original framing imagined a broader "extract all mechanical sequences" effort with packaging, shipping, and override concerns. After discussion, narrowed to: move skills/* → skills/cc/*, add skills/cc/scripts/sync-base-branch.py, wire three call-sites (/start Case A Step 4, /start Case B Step 4, /deploy Step 1).

Deferred because the skills/cc/ move requires sync.py layout awareness that #157 will deliver. Landing this first would ship the project in a broken intermediate state and entangle any post-#157 bugs with this work. Pick this back up once #157 has merged.

Subsequent extractions (--body-file plumbing, follow-up-issue loops, label/milestone resolution) are explicitly out of scope here and should get their own tickets after this tracer bullet validates the pattern.

[sebastientaggart]

Resolution

The repetitive multi-line "make sure the base branch is clean and matches the remote exactly" step that lived inside /start and /deploy is now a single command — a small Python helper that does the work in one shot. This makes the workflow noticeably less chatty (no more permission prompts on each piece of the sequence), more reliable (one atomic operation per run, with clear error messages if the working tree isn't clean), and easier to extend the same pattern to other repetitive bits later.

See #162 for full technical details.