@curie-agent/core 0.2.5 → 0.3.3

package/templates/skills/deep-research/references/report-structure.md

@@ -0,0 +1,149 @@
+# Deep Report Structure Template
+
+Use this template when the user asks for a comprehensive report, white paper, literature review,
+or any long-form research output. Adapt sections as needed — not every section applies to every topic.
+
+---
+
+## Report Template
+
+```markdown
+# [Title: Clear, Specific, Descriptive]
+
+**Prepared for**: [Audience / purpose]
+**Date**: [Month Year]
+**Research scope**: [Time period covered, geographic scope, source types used]
+
+---
+
+## Executive Summary
+
+[3–5 sentences maximum. Answer: What was researched, what are the key findings, and what are
+the main implications. This is the only section many readers will read — make it count.]
+
+---
+
+## Introduction
+
+**Background**: [Why this topic matters / context]
+**Research questions**: [The specific questions this report addresses]
+**Scope and limitations**: [What is and isn't covered, and why]
+**Methodology**: [How research was conducted: source types, search strategy, evaluation criteria]
+
+---
+
+## Section 1: [Core Topic / Foundational Context]
+
+[Establish what is known, agreed upon, and well-documented. This section grounds the reader.]
+
+### Key findings
+- ...
+
+### Sources basis
+[Note what types of sources underpin this section and their quality]
+
+---
+
+## Section 2: [Main Analysis / Findings]
+
+[The substantive core of the research. Use sub-sections as needed.]
+
+### 2.1 [Sub-topic]
+...
+
+### 2.2 [Sub-topic]
+...
+
+---
+
+## Section 3: [Debates, Disagreements, Uncertainty]
+
+[Where experts or sources disagree. Be fair to all serious positions.]
+
+### Area of disagreement 1
+**Position A**: [What some argue, with evidence]
+**Position B**: [What others argue, with evidence]
+**Assessment**: [What the weight of evidence suggests, or note if genuinely unresolved]
+
+---
+
+## Section 4: [Implications / So What?]
+
+[What do these findings mean in practice? Connect back to why the user needed this research.]
+
+---
+
+## Section 5: [Gaps & Future Directions] *(optional)*
+
+[What isn't well-studied? What questions remain open? What should the reader watch for?]
+
+---
+
+## Conclusions
+
+[Synthesize: What does the body of evidence say? What is the answer to the main question,
+with appropriate confidence? What are the most important caveats?]
+
+---
+
+## Recommendations *(if applicable)*
+
+[Only include if the user asked for actionable guidance. Base firmly on the evidence above.
+Use hedged language: "The evidence suggests...", "Consider...", "If X is true, then..."]
+
+---
+
+## Sources
+
+[Annotated bibliography — not just URLs. For each key source, include:]
+- Full citation (author, title, publication, date, url)
+- Brief note: what it contributes and any relevant caveats (e.g., industry-funded, older study)
+
+Group by: Primary sources | Secondary analysis | Background context
+
+---
+
+## Appendix *(optional)*
+
+[Data tables, detailed methodology notes, extended quotes, glossary of technical terms]
+```
+
+---
+
+## Length Guidelines
+
+| Report Type | Approximate Length |
+|---|---|
+| Quick brief | 300–600 words |
+| Standard summary | 600–1,200 words |
+| Research report | 1,200–3,000 words |
+| Comprehensive deep dive | 3,000–6,000 words |
+| Literature review / white paper | 5,000+ words |
+
+Always ask the user their preference if not specified — defaulting to the shortest format that fully answers the question is usually best.
+
+---
+
+## Tone and Style Calibration
+
+Adjust writing to the user's context:
+
+| Audience | Tone | Jargon level | Citation style |
+|---|---|---|---|
+| Personal / curious | Conversational, engaging | Low — explain terms | Informal ("According to a 2023 Harvard study...") |
+| Business professional | Clear, direct, executive | Medium | Inline with source name |
+| Academic | Formal, precise | High (field-appropriate) | APA / MLA / Chicago |
+| Journalist | Accessible, vivid | Low-medium | Clear attribution |
+| Student | Thorough, explanatory | Medium | As required by assignment |
+
+---
+
+## Common Report Mistakes to Avoid
+
+- **Padding**: length without substance — every paragraph should add something
+- **Missing the forest for the trees**: lots of facts, no synthesis or conclusion
+- **False certainty**: presenting contested findings as settled
+- **Burying the lede**: the most important finding should not be on page 4
+- **Source monoculture**: citing only one type of source (e.g., only news, only studies)
+- **Recency bias**: assuming recent = better without checking quality
+- **No practical bridge**: research without connecting to why the user needs it

package/templates/skills/deep-research/references/topic-guides.md

@@ -0,0 +1,151 @@
+# Topic-Specific Research Guides
+
+Reference this file when conducting research in specialized domains.
+
+---
+
+## Medical & Health Research
+
+### Best Sources
+- PubMed / MEDLINE — indexed medical research papers
+- Cochrane Library — systematic reviews and meta-analyses (highest quality evidence)
+- WHO, CDC, NIH, NHS — institutional health guidance
+- Major medical journals: NEJM, Lancet, JAMA, BMJ
+
+### Evidence Hierarchy (strongest → weakest)
+1. Systematic reviews & meta-analyses of randomized trials
+2. Randomized controlled trials (RCTs)
+3. Cohort studies and observational research
+4. Expert consensus / clinical guidelines
+5. Case reports, anecdotes, expert opinion
+
+### Key Warnings
+- Single studies (even published ones) can be wrong or misleading — look for replication
+- "Correlation ≠ causation" — observational studies can only suggest links
+- Supplement and diet research is often industry-funded — check for conflicts of interest
+- Relative risk vs absolute risk: "50% more risk" sounds alarming; "0.001% → 0.0015%" is not
+- Always note: individual medical situations vary — encourage consulting a professional for personal health decisions
+
+### Research Pattern
+Search: specific condition/treatment → look for systematic reviews → check professional guidelines → note disagreements between guidelines → identify areas of uncertainty
+
+---
+
+## Legal & Regulatory Research
+
+### Best Sources
+- Official government websites (legislation, court rulings, regulatory agencies)
+- Legal databases: Google Scholar (case law), court official websites
+- Bar association publications
+- Academic law journals for analysis
+
+### Key Principles
+- **Jurisdiction matters enormously**: law varies by country, state/province, city
+- **Distinguish**: statute (written law) vs case law (court interpretations) vs regulation (agency rules)
+- **Currency**: laws change — always note when you found the information
+- Emphasize strongly that specific legal situations require a qualified lawyer
+
+### Research Pattern
+Identify jurisdiction → find applicable statute or regulation → find court interpretations → check for recent amendments or cases → note if the law is settled or contested
+
+---
+
+## Financial & Market Research
+
+### Best Sources
+- SEC filings (EDGAR) for public company data in the US
+- Company annual reports and investor relations pages
+- Central bank publications (Federal Reserve, ECB, etc.)
+- IMF, World Bank for macroeconomic data
+- Established financial news: FT, WSJ, Bloomberg, Reuters
+- Industry analyst reports (note: often have buy/sell conflicts)
+
+### Key Warnings
+- Financial forecasts are often wrong — present ranges, not certainties
+- Distinguish: historical data (fact) vs projections (opinion/model)
+- Be explicit about who funded industry reports
+- Investment advice requires professional licensing — don't make specific recommendations; present information
+
+### Research Pattern
+Macro context → industry overview → company/asset specifics → historical performance → risks and counterarguments → expert opinion range
+
+---
+
+## Scientific & Academic Research
+
+### Best Sources
+- Google Scholar, Semantic Scholar, arXiv (preprints)
+- PubMed, IEEE Xplore (by discipline)
+- University institutional repositories
+- Peer-reviewed journals in the specific field
+
+### Key Principles
+- **Preprints** (arXiv, bioRxiv) are not peer-reviewed — flag this clearly
+- **Peer review is a floor, not a ceiling** — published studies still get retracted or contradicted
+- Look for **replication**: single studies that haven't been replicated warrant caution
+- **Effect size** matters as much as statistical significance
+- Note the study's funding source and any potential conflicts of interest
+
+### Research Pattern
+Find seminal/foundational papers → find recent systematic reviews → identify ongoing debates in the literature → look for consensus statements from scientific bodies → note open questions
+
+---
+
+## Historical Research
+
+### Best Sources
+- Academic history books and journal articles
+- National archives and libraries
+- Digitized primary sources: newspapers, official documents, letters
+- Encyclopedias for established facts (Wikipedia as a starting point, not endpoint)
+- Oral history archives for recent history
+
+### Key Principles
+- **Primary sources** (documents from the time) vs **secondary sources** (historians' analysis) — both are valuable, serve different purposes
+- History is interpreted differently across time and perspective — note historiographical debates
+- Be specific about dates, places, and actors — vagueness is a red flag in historical claims
+- Distinguish between established historical consensus and revisionist interpretations
+
+### Research Pattern
+Establish basic chronology → identify key actors and causes → find primary source evidence → read historians' interpretations → note where scholars disagree → contextualize in broader history
+
+---
+
+## Investigative & Journalistic Research
+
+### Best Sources
+- Investigative journalism outlets: ProPublica, The Intercept, Reuters Investigates
+- Public records, FOIA documents
+- Court filings (PACER in the US, court websites)
+- Non-profit watchdog organizations
+- Academic researchers in relevant fields
+
+### Key Principles
+- **Document everything**: note where every fact came from
+- **Two-source rule**: important claims need at least two independent sources
+- Distinguish **on record** (citable) vs **background** (contextual, not attributed)
+- Look for **conflicts of interest** in every source — who funds them? What's their stake?
+- **Follow the money**: financial records often reveal motivations
+
+### Research Pattern
+Define the specific claim or question to investigate → find documentary evidence first → identify relevant experts and officials → search for contradicting evidence → assess source motivations → build a chain of evidence
+
+---
+
+## Technology Research
+
+### Best Sources
+- Official documentation (for products, standards, APIs)
+- Peer-reviewed computer science papers (arXiv, ACM, IEEE)
+- Reputable tech journalism: Ars Technica, The Verge, Wired (for general), specialized outlets
+- Company blogs (treat as primary source with obvious bias)
+- GitHub repositories for open source (community reaction tells you a lot)
+
+### Key Warnings
+- Tech moves fast — check publication dates carefully
+- Vendor materials are marketing — cross-reference with independent reviews
+- Benchmark results are highly context-dependent
+- "AI can do X" claims often require careful reading of what "X" actually means in context
+
+### Research Pattern
+Understand the technology category → find official specifications → find independent testing/reviews → understand limitations and failure modes → compare alternatives → check adoption and community health

package/templates/skills/planning/SKILL.md

@@ -0,0 +1,222 @@
+---
+name: planning
+description: >
+  Use this skill for any kind of planning task — personal, professional, creative, or logistical.
+  This covers a broad range: project planning, life planning, event planning, travel planning,
+  business strategy, goal setting, habit building, content calendars, study plans, home renovation
+  timelines, wedding planning, meal planning, financial planning, career planning, and more.
+  Trigger this skill whenever the user wants to organize actions over time, structure a complex
+  goal into steps, create a roadmap, break down an overwhelming task, prioritize competing demands,
+  or build a structured schedule or timeline. Trigger even if the request seems casual —
+  "help me plan my week", "I want to launch a course, where do I start", "how should I approach
+  renovating my kitchen", "plan a birthday party for 30 people", "I'm overwhelmed, help me organize
+  my life" — all of these should use this skill. This is NOT limited to software development.
+---
+
+# Planning Skill
+
+A universal skill for helping users create clear, actionable, and realistic plans across any domain of life or work.
+
+---
+
+## Core Philosophy
+
+Great planning has three enemies: **vagueness**, **overwhelm**, and **rigidity**. Your job is to:
+1. Make goals concrete and outcomes measurable
+2. Break work into steps small enough to actually start
+3. Build in flexibility so the plan survives contact with reality
+
+Always anchor the plan to the **person's actual context** — their time, energy, resources, constraints, and priorities — not an idealized version.
+
+---
+
+## Step 0: Understand Before Planning
+
+Before writing a single bullet point, gather enough context to plan well. The key questions:
+
+| Question | Why It Matters |
+|---|---|
+| What is the end goal? | Without a destination, you can't map a route |
+| What does success look like? | Helps define milestones and "done" |
+| What's the timeline / deadline? | Determines pace and tradeoffs |
+| What resources are available? | Time, money, people, tools, energy |
+| What constraints exist? | Budget limits, non-negotiables, dependencies |
+| What's already been done? | Avoid re-planning what's settled |
+| What's the biggest risk or unknown? | Informs where to build in buffers |
+
+If you're missing critical info, **ask first** — one focused question at a time. Don't write a plan for the wrong goal.
+
+---
+
+## Planning Modes
+
+Choose the right mode based on what the user needs. Often you'll combine modes.
+
+### 1. 🗺️ Roadmap Planning
+*Use when: the user has a big goal with no clear path yet.*
+
+- Define the end state clearly
+- Work backwards to identify major milestones
+- Group milestones into phases (e.g., Foundation → Build → Launch → Sustain)
+- Assign rough timeframes to each phase
+- Identify dependencies between milestones
+
+**Output format**: Phased roadmap with milestones and timeframes. See `references/templates.md → Roadmap`.
+
+---
+
+### 2. 📅 Timeline / Schedule Planning
+*Use when: the user has a deadline and needs to fit work into available time.*
+
+- Anchor to the deadline and work backwards
+- Identify fixed dates (events, appointments, dependencies)
+- Estimate time for each task honestly — add 20–30% buffer
+- Assign tasks to specific days/weeks
+- Flag overloaded periods proactively
+
+**Output format**: Week-by-week or day-by-day schedule. See `references/templates.md → Timeline`.
+
+---
+
+### 3. 🎯 Goal Decomposition
+*Use when: the user has a goal but is overwhelmed and doesn't know where to start.*
+
+- Restate the goal clearly in concrete terms
+- Break it into 3–7 major outcomes needed
+- For each outcome, list 3–5 specific actions
+- Identify the very **first action** — make it small enough to do today
+- Surface blockers and unknowns upfront
+
+**Output format**: Goal → Outcomes → Actions tree. See `references/templates.md → Decomposition`.
+
+---
+
+### 4. 🏗️ Project Planning
+*Use when: the user is managing a multi-person or multi-phase effort with deliverables.*
+
+- Define scope: what's IN and OUT of this project
+- List all deliverables
+- Map dependencies between tasks
+- Assign ownership (if relevant)
+- Set checkpoints / review points
+- Define risks and mitigations
+
+**Output format**: Project brief + task list with owners and dates. See `references/templates.md → Project`.
+
+---
+
+### 5. 📆 Periodic Planning (Weekly / Monthly / Quarterly)
+*Use when: the user wants to organize recurring time periods.*
+
+- Review what happened last period (if known)
+- Surface priorities and commitments for this period
+- Allocate time blocks to categories (work, personal, health, relationships, etc.)
+- Set 1–3 "wins" to aim for — not a perfect list, just top priorities
+- Leave margin for the unexpected
+
+**Output format**: Time-block schedule + top priorities list. See `references/templates.md → Periodic`.
+
+---
+
+### 6. 🎉 Event / Experience Planning
+*Use when: the user is organizing an event, trip, celebration, or one-time experience.*
+
+- Define: Who, What, When, Where, How many people, Budget
+- Break into: Pre-event tasks | Day-of tasks | Post-event tasks
+- Create vendor/supplier list if needed
+- Build a communications plan (invites, reminders, follow-ups)
+- Identify what can go wrong and have a backup
+
+**Output format**: Event checklist with timeline. See `references/templates.md → Event`.
+
+---
+
+### 7. 🔄 Habit & Routine Planning
+*Use when: the user wants to build a sustainable system, not a one-time plan.*
+
+- Anchor new habits to existing routines (habit stacking)
+- Start smaller than feels necessary — consistency > intensity
+- Design the environment to reduce friction
+- Define a tracking method
+- Build in a weekly review moment
+
+**Output format**: Routine design + habit tracker structure. See `references/templates.md → Habit`.
+
+---
+
+## Universal Planning Principles
+
+Apply these regardless of the mode:
+
+### Prioritization
+When there's more to do than time allows, help the user prioritize. Use the **Impact × Urgency** lens:
+- High impact + urgent → Do first
+- High impact + not urgent → Schedule deliberately (these are often neglected)
+- Low impact + urgent → Delegate or batch
+- Low impact + not urgent → Drop or defer
+
+### Buffer and Realism
+- Most people underestimate time by 30–50%
+- Build recovery time after intense periods
+- Mark "must haves" vs "nice to haves" explicitly
+
+### Dependencies
+- Identify what must happen *before* something else can start
+- Surface these early — they're where plans collapse
+
+### Energy, Not Just Time
+- Help the user match task type to their energy level
+- Deep focus work → morning (for most people)
+- Admin, email, calls → afternoon
+- Creative / brainstorming → after movement or meals
+
+### The First Next Action
+Always end with: *"What's the very first concrete action you'll take?"* This closes the loop between planning and doing.
+
+---
+
+## Output Quality Standards
+
+A good plan output:
+- Is **specific** (not "work on the project" but "draft the introduction section")
+- Is **time-bound** (has dates, durations, or deadlines)
+- Is **realistic** (accounts for the user's actual schedule)
+- Is **actionable** (the next step is clear without more planning)
+- Is **appropriately detailed** (not so granular it's overwhelming, not so vague it's useless)
+- Has a **clear visual structure** (tables, phases, or numbered lists depending on content)
+
+---
+
+## Domain-Specific Guidance
+
+For specialized domains, read the relevant reference file before planning:
+
+| Domain | Reference File |
+|---|---|
+| Business / Startup / Product | `references/business.md` |
+| Personal Life / Wellbeing / Career | `references/personal.md` |
+| Events / Travel / Experiences | `references/events.md` |
+| Learning / Education / Skill Building | `references/learning.md` |
+| Creative Projects (writing, art, content) | `references/creative.md` |
+
+---
+
+## Presenting the Plan
+
+- Use headers to separate phases or categories
+- Use tables for schedules and timelines
+- Use numbered lists for sequences where order matters
+- Use bullet lists for non-ordered items
+- Bold key dates, owners, and critical-path items
+- Always end with a "Next Steps" or "Start Here" section
+- Offer to adjust, drill down on a phase, or export as a document
+
+---
+
+## Anti-Patterns to Avoid
+
+- ❌ **The Perfect Plan Trap**: Don't over-engineer. A 70% good plan that gets started beats a 100% plan that never does.
+- ❌ **Planning Without Constraints**: Never ignore time, money, or energy limits — they're the whole game.
+- ❌ **Front-loading Detail**: Don't plan week 6 in detail when week 1 is still unclear. Plan near-term in detail, far-term in rough strokes.
+- ❌ **Missing the Human**: Always consider motivation, fear, habit, and energy — not just tasks and dates.
+- ❌ **No Review Cadence**: Every plan needs a moment to check in, adapt, and course-correct.

package/templates/skills/planning/references/business.md

@@ -0,0 +1,82 @@
+# Business, Startup & Product Planning
+
+Domain-specific guidance for planning in business, entrepreneurship, and product contexts.
+
+---
+
+## Business Planning Areas
+
+### Business / Startup Planning
+Key questions:
+- What problem does this solve and for whom?
+- What's the business model (how does it make money)?
+- What's the competitive landscape?
+- What resources are needed to get to first revenue?
+- What does the founding team look like?
+
+Common phases:
+1. **Validation** — Prove the problem is real and people will pay
+2. **MVP** — Build the minimum that delivers value
+3. **Launch** — Get first paying customers
+4. **Growth** — Replicate and scale what works
+5. **Scale** — Systematize, hire, expand
+
+Use a **lean planning approach**: 1-page plan, test assumptions fast, update often.
+
+### Product Planning
+Key inputs needed:
+- Who is the user / customer?
+- What job are they trying to do?
+- What are the top 3 user problems to solve?
+- What's already been built vs. what's being planned?
+- What's the release cadence (continuous / quarterly / etc.)?
+
+Frameworks to apply:
+- **Now / Next / Later roadmap**: Avoid fake precision on distant quarters
+- **User story mapping**: Arrange features by user journey and slice by MVP
+- **RICE scoring**: Reach × Impact × Confidence ÷ Effort for prioritization
+
+### Marketing & Launch Planning
+Structure:
+1. Define target audience (be specific — not "everyone")
+2. Set launch goal (signups, revenue, press coverage, etc.)
+3. Map the customer journey (awareness → consideration → decision)
+4. Assign channels (organic, paid, email, partnerships, PR)
+5. Create content calendar
+6. Define success metrics and review cadence
+
+### Team & Hiring Planning
+- Map current gaps to goals (what roles are missing?)
+- Sequence hires by impact (what hire unlocks the most?)
+- Build a realistic hiring timeline (3–4 months per senior hire)
+- Plan for onboarding time (new hire ≠ productive hire immediately)
+
+---
+
+## Business Planning Frameworks
+
+### SWOT (quick strategic lens)
+- Strengths: What advantages do you have?
+- Weaknesses: Where are you vulnerable?
+- Opportunities: What external factors could you exploit?
+- Threats: What could undermine you?
+
+Use SWOT for orientation, not as a plan itself.
+
+### North Star Metric
+Pick the ONE metric that best captures the value you deliver to users. Every plan should explain how it moves this metric.
+
+### Working Backwards (Amazon-style)
+Start with the ideal press release announcing the finished product/outcome. Then figure out what needs to be true to make it real.
+
+### Stage-Gate Planning
+Define clear gates between phases. A project only moves to the next stage when it passes specific criteria (e.g., "10 paying customers" before building v2).
+
+---
+
+## Business Planning Anti-Patterns
+
+- **Planning without validating assumptions**: Know which assumptions your plan depends on and test the risky ones first
+- **Activity plans vs. outcome plans**: "Post 3x per week" is activity. "Reach 1,000 newsletter subscribers" is outcome. Plan for outcomes.
+- **Ignoring cash flow timing**: Especially for early-stage businesses — revenue on paper ≠ cash in hand
+- **Planning in isolation**: Business plans need to account for how customers, competitors, and team will actually behave