Agentic Coding

**Verdict:** A strong, process-heavy skill for teams that want merge-ready, reviewer-friendly AI-generated code with explicit quality gates and local-only artifacts. Best suited to production work, not quick experiments. **What it does well** - Imposes a **contract-first workflow** (objective, acceptance, non-goals, constraints) that sharply reduces ambiguity and scope drift. - Uses the **PACT loop** to normalize how changes are framed, validated, and documented. - Prioritizes **small, surgical diffs** and discourages broad rewrites, improving reviewability and rollback safety. - For bugs, demands **proof of failure before fix**, plus after-the-fact evidence, reducing unverifiable “fixed” states. - Generates **handoff packets** (changes, files touched, validation results, risks, rollback path) that fit well into serious review processes. - Stores contracts, evidence, and handoffs **locally in `~/agentic-coding/`** and explicitly avoids undeclared network access. **Key risks / limitations** - **Process overhead:** The strict contract and PACT loop can feel heavy for small tasks, spikes, or exploratory coding—this is not optimized for rapid, informal iteration. - **User discipline required:** Reliability depends on you actually running tests/commands and maintaining the local memory files; the skill enforces structure, not truth of external systems. - **Narrow scope by design:** It focuses on execution quality and handoff, not ideation, architecture exploration, or multi-agent orchestration (those are delegated to related skills). - **Potential friction with existing workflows:** Teams with informal or ad-hoc review norms may initially find the structure rigid or redundant. **Recommended scenarios** - You want an AI agent to produce **merge-ready code** with clear acceptance criteria and minimal, reviewable diffs. - You’re handling **production features, risky refactors, or release-branch hotfixes** where rollback and evidence matter. - You work heavily in **Xcode/Swift/iOS/macOS** and need the AI to align to concrete targets, devices, and test commands. - You’re integrating AI into a **serious engineering process** and want explicit contracts, verification snapshots, and deterministic handoffs instead of free-form coding assistance.