ONE REPOSITORY
FOR THE ENTIRE PROMPT STACK.

Reviews a PR like a senior who optimizes for the next reader, not style nits.

Write an ADR that future-you will still understand in 18 months.

Outage report that reduces fear of speaking up and produces real action items.

Reach a busy founder without AI-slop tone. Four lines, one ask, no hedging.

Monthly update that earns trust by being honest about misses.

Announce a B2B SaaS launch in the YC style. Sharp lede, two metrics, one founder quote.

Ask the questions that surface pain, not features. Seven questions, no leading the witness.

Set pricing for a new B2B SaaS. Three tiers, unit economics, anchor + decoy logic.

Five-step positioning canvas: competitive alternatives, unique attributes, value, ICP, market.

Distill 12 customer calls into themes, frequencies, and a top-3 list that survives committee.

Hook earns the open. 3 curated items, one personal note, soft CTA.

Hook → reframe → 3 acts → payoff. Designed to hold 60%+ retention.

Announce a feature ship. Lead with the job it does, not the feature name.

Submit a talk that reviewers can't skip. Hook, claim, 3 takeaways, audience.

Compress a 30-page paper into 5 lines + 4 structured takeaways.

Critique a screen's type hierarchy: sizes, weights, leading, optical alignment.

An empty state that teaches the user what to do next. Not just a blank screen.

Error messages that respect the user. Specific, actionable, never condescending.

Acknowledge, explain, propose, follow up. Don't roll over.

Reply to a SIG Lite questionnaire using your SOC 2 report. Cite controls.

Take wishful OKRs and turn them into bets you can actually grade.

Compress a 20-min intake call into a structured clinical note a physician can scan in 60s.

Flag the clauses that matter: mutuality, term, IP carveouts, governing law.

Teach a concept by asking, not telling. Five questions ordered by difficulty.

Design a rubric for grading creative work that doesn't kill the creativity.

Write a policy doc (e.g., remote work, AI usage) that's clear, lived, and won't gather dust.

Sell without lying. Lead with the JTBD. Three specs. Risk reversal.

Sell the moment, not the square footage. 3 paragraphs.

Write a listing that wins the algorithm AND the buyer. Keywords, bullets, A+ copy.

Decision memo that earns the meeting. Context, options, recommendation, risks.

Build a board deck that tells one story. Not 40 slides of metrics.

Design a 5-stage interview loop that surfaces real signal, not theater.

Have a hard performance conversation that respects both people. Script + recovery plan.

Build a 10-slide deck investors don't skim. Inevitability, wedge, traction, ask.

Run a pre-mortem on a decision. Imagine it's failed; trace back the most likely cause.

Build a system prompt for a custom agent that doesn't drift. Role, constraints, format.

Deliver feedback that the recipient can hear AND act on. Specific, kind, future-focused.

Design an A/B test that's powered, falsifiable, and shippable. Hypothesis to readout.

Argue a contrarian position credibly. Steelman opposition, then dismantle.

5-email onboarding that gets users to first value, not just first login.

Write a spec that fits on one page and ships in one sprint.

Diagnose why a prompt is producing poor output and fix it.

Designs a complete, production-grade system prompt for an autonomous agent including persona, tool contracts, guardrails, and stop conditions.

Builds a strict ReAct-style Thought/Action/Observation loop with explicit formatting and self-correction rules for tool-using agents.

Plans a coordinated multi-agent team with roles, hand-off contracts, shared memory, and conflict resolution for a complex objective.

Writes precise tool/function definitions with JSON schemas, descriptions, and usage hints that minimize wrong-tool and bad-argument errors.

Adds a rigorous self-review pass where the agent critiques its own draft action against a checklist before committing it.

Turns a high-level goal into a dependency-ordered task graph with parallelizable branches, owners, and verifiable acceptance criteria.

Directs a retrieval-augmented agent to query, ground answers in sources, cite spans, and refuse when evidence is insufficient.

Defines a structured error-handling policy covering retries, backoff, alternative tools, and graceful degradation for agent runtimes.

Specifies where an autonomous agent must pause for human approval, what to present, and how to resume after a decision.

Designs short-term, long-term, and episodic memory plus a summarization policy to keep an agent coherent across long sessions.

Drives a web-browsing agent to navigate, extract, and act on pages reliably with verification and anti-hallucination guards.

Creates an objective rubric and grades an agent execution trace on task success, tool use, efficiency, and safety.

Compiles a layered guardrail set covering input validation, action allow-lists, output filtering, and prompt-injection defense.

Specifies a recurring agent that monitors a signal, decides if action is warranted, and reports or escalates on a schedule.

Generates a structured battery of prompt-injection and jailbreak attacks to stress-test an agent before deployment.

Analyzes an agent workflow and proposes concrete changes to cut token cost and step count without losing task quality.

Runs an autonomous support agent that classifies tickets, resolves what it can, and escalates the rest with full context.

Governs a coding agent to explore, plan, implement in small verifiable steps, and self-test before declaring done.