gemini-prompting

Gemini Prompting

Use this skill when gemini:gemini-rescue needs to ask Gemini for help.

Prompt Gemini 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 Gemini run. Split unrelated asks into separate runs.
• Tell Gemini 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.
• Prefer better prompt contracts over raising reasoning or adding long natural-language explanations.
• Use XML tags consistently so the prompt has stable internal structure.
• Gemini 2.5 Pro and Flash are strong at long-context code edits and tool use but reward explicit contracts. Spell out file boundaries, acceptance criteria, and stop conditions.

Default prompt recipe:

• <task>: the concrete job and the relevant repository or failure context.
• <structured_output_contract> or <compact_output_contract>: exact shape, ordering, and brevity requirements.
• <default_follow_through_policy>: what Gemini should do by default instead of asking routine questions.
• <verification_loop> or <completeness_contract>: required for debugging, implementation, or risky fixes.
• <grounding_rules> or <citation_rules>: required for review, research, or anything that could drift into unsupported claims.

When to add blocks:

• Coding or debugging: add completeness_contract, verification_loop, and missing_context_gating.
• Review or adversarial review: add grounding_rules, structured_output_contract, and dig_deeper_nudge.
• Research or recommendation tasks: add research_mode and citation_rules.
• Write-capable tasks: add action_safety so Gemini stays narrow and avoids unrelated refactors.

How to choose prompt shape:

• Use the dedicated /gemini:review command to invoke Gemini's built-in /review reviewer for whole-diff reviews. Use /gemini:adversarial-review for the strict, JSON-schema'd adversarial reviewer.
• Use task when the task is diagnosis, planning, research, or implementation and you need to control the prompt directly.

Working rules:

• Prefer explicit prompt contracts over vague nudges.
• Use stable XML tag names.
• Do not raise reasoning or complexity first. Tighten the prompt and verification rules before escalating.
• Ask Gemini for brief, outcome-based progress updates only when the task is long-running or tool-heavy.
• Keep claims anchored to observed evidence. If something is a hypothesis, say so.

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 whether Gemini should keep going by default or stop for missing high-risk details.
4. Add verification, grounding, and safety tags only where the task needs them.
5. Remove redundant instructions before sending the prompt.

Reusable blocks

<task>
  {{concrete job, including file paths, failure messages, or repo state}}
</task>

<structured_output_contract>
  Return a Markdown document with this exact outline:
  ## Summary
  - One sentence.
  ## Findings
  - [severity] title — file:line — one-sentence detail.
  ## Fix plan
  - Ordered list of concrete steps.
  Keep each bullet under 160 characters. No prose outside these sections.
</structured_output_contract>

<default_follow_through_policy>
  If a step is blocked, document the blocker and continue with the next step
  rather than asking a clarifying question. Only stop for missing high-risk
  information (data-loss, auth, destructive migration).
</default_follow_through_policy>

<verification_loop>
  After editing, run the project's test command (`npm test` or the command
  listed in package.json) and paste the last 40 lines of output under a
  `## Verification` heading. If tests fail, iterate until they pass or you
  hit the same failure twice — then stop and report.
</verification_loop>

<grounding_rules>
  Every finding must cite a concrete file path and line range. If a claim
  is an inference, prefix it with "Inference:" and cite the evidence.
</grounding_rules>

<action_safety>
  Touch only files directly related to the task. Do not reformat, rename,
  or refactor unrelated code. If you must edit more than 5 files, stop and
  list them in a `## Requested scope expansion` block first.
</action_safety>