claude-prompting

Claude Prompting

Use this skill when the claude-rescue flow needs to hand a task to Claude Code.

Prompt Claude like an operator, not a collaborator. Keep prompts compact and block-structured with XML tags. State the task, the output contract, the follow-through defaults, and the small set of extra constraints that matter.

Core rules:

• Prefer one clear task per Claude run. Split unrelated asks into separate runs.
• Tell Claude what done looks like. Do not assume it will infer the desired end state.
• Add explicit grounding and verification rules for any task where unsupported guesses would hurt quality.
• Claude explores the repository itself; describe the goal and constraints, not a file-by-file tour.
• Use XML tags consistently so the prompt has stable internal structure.

Default prompt recipe:

• <task>: the concrete job and the relevant failure or feature context.
• <output_contract>: exact shape, ordering, and brevity requirements for the final answer.
• <default_follow_through_policy>: what Claude should do by default instead of asking routine questions (headless runs cannot ask).
• <verification_loop>: required for debugging, implementation, or risky fixes — name the commands Claude should run to prove the change works.
• <grounding_rules>: required for review, research, or anything that could drift into unsupported claims.

When to add blocks:

• Coding or debugging: add verification_loop and a completeness requirement ("fix every call site, not just the reported one").
• Review-like asks: use the dedicated review/adversarial-review subcommands instead; their prompts already carry the review contract.
• Planning: use the plan subcommand; implementation: use implement.
• Write-capable tasks: add an action_safety block so Claude stays narrow and avoids unrelated refactors.

How to choose prompt shape:

• Use task --resume-last for follow-up instructions on the same Claude session. Send only the delta instruction instead of restating the whole prompt unless the direction changed materially.
• Remember headless constraints: Claude cannot ask clarifying questions mid-run. Resolve ambiguity in the prompt or tell Claude which assumption to make.

Prompt assembly checklist:

1. Define the exact task and scope in <task>.
2. Choose the smallest output contract that still makes the answer easy to use.
3. Decide what Claude should do when it hits a missing detail: proceed with a stated assumption, or stop and report.
4. Add verification, grounding, and safety tags only where the task needs them.
5. Remove redundant instructions before sending the prompt.

Reusable blocks live in references/prompt-blocks.md. Concrete end-to-end templates live in references/claude-prompt-recipes.md. Common failure modes to avoid live in references/claude-prompt-antipatterns.md.