Prompt library

Prompts that favor clear outcomes.

Small specs beat vague wishes. Use these cards as starting points, then tune the details for your context.

Details change the answer. A strong prompt reads like a tiny spec: it states the goal, the audience, the constraints, the inputs you have, and what "good" looks like. The cards below are starting points. Copy/paste them into your favorite text editor, swap the placeholders for your details, and prune the sections you do not need. Then copy/paste that into your LLM chat of choice.

Clarify a vague request

Turn a fuzzy ask into questions, assumptions, and a plan before acting.

task (required) constraints (optional) audience (optional)
You are a careful facilitator. Someone asked: {{task}}
Constraints to respect: {{constraints}}
Audience: {{audience}}

Do the following:
1) Restate the task in one precise sentence.
2) List the critical details that are missing or ambiguous.
3) Ask up to 5 targeted questions that would remove the ambiguity.
4) Offer a short draft plan (3-6 steps) that would satisfy the request, calling out any assumptions.
5) Stop and wait for answers before executing the plan.

Code change plan from repo map

Produce a surgical change plan before touching code.

feature_goal repo_map (directories, key files, languages) quality_checks (tests, linters, reviews)
You are drafting a change plan for a codebase.
Feature goal: {{feature_goal}}
Repository map: {{repo_map}}
Quality checks to keep: {{quality_checks}}

Deliver:
- The likely files or modules to edit (cite why each matters).
- A numbered implementation plan with tight, verifiable steps.
- Risks or unknowns that could derail the change.
- A test and validation plan covering happy path, edge cases, and failure modes.
Keep it concise; avoid generic advice.

Hypothesis-first log triage

Turn noisy logs into the next investigative steps.

symptoms (what the user sees) log_excerpt (timestamped lines) recent_changes (deploys, config flips)
Act like a disciplined SRE triaging an incident.
Symptoms reported: {{symptoms}}
Recent changes: {{recent_changes}}
Log excerpt: {{log_excerpt}}

Provide:
- 2-4 hypotheses ranked by likelihood; include the signals that support each.
- The smallest next check to confirm or reject each hypothesis.
- A quick, safe mitigation to reduce user impact while investigating.
Avoid speculation without citing the provided data.

Decision log from meeting notes

Extract decisions and owners instead of a wall of text.

raw_notes attendees (optional) deadlines (if mentioned)
Convert the meeting notes into a crisp decision log.
Notes: {{raw_notes}}
Attendees: {{attendees}}
Deadlines heard: {{deadlines}}

Output a markdown list with:
- Decisions (what, owner, due date if present).
- Open questions (what is needed, who owes it, by when).
- Follow-up actions (owner, expected outcome, checkpoint date).
Keep wording factual and avoid inventing owners or dates.