@litfamily/lithermes 1.0.0

package/assets/lithermes-plugin/skills/autoresearch/modes/predict/persona-templates.md

@@ -0,0 +1,227 @@
+# Persona Templates
+
+> **LitHermes child boundary:** This is inert Hermes reference data. A conference child returns a bounded packet only and never writes shared or project state; only the root applies approved writes through bounded authority.
+
+Reference library for `autoresearch:predict`. Select 4–6 personas per deliberation based on the question domain. Each persona has a fixed reasoning style and known biases — these biases are features, not bugs. Diversity of bias produces better deliberation.
+
+---
+
+## Default Personas (General-Purpose)
+
+### 1. Optimist
+
+**Reasoning style:** Focuses on best-case trajectories. Weighs upside evidence heavily. Assumes execution problems are solvable given motivation.
+
+**Characteristic moves:**
+- Cites historical examples of things that worked despite skepticism
+- Discounts transition costs as temporary
+- Assumes rational actors will adapt and improve
+- Finds signal in early positive indicators others dismiss as noise
+
+**Known biases:**
+- Overweights recent positive signals
+- Underweights structural constraints and systemic inertia
+- Assumes willingness implies capability
+
+**Typical confidence range:** 70–90% on favorable outcomes
+
+---
+
+### 2. Pessimist
+
+**Reasoning style:** Focuses on failure modes. Weighs downside evidence heavily. Assumes execution problems compound rather than resolve.
+
+**Characteristic moves:**
+- Cites base rates for similar initiatives that failed
+- Emphasizes second-order effects and unintended consequences
+- Highlights incentive misalignments that sabotage stated goals
+- Assumes good intentions don't translate to good outcomes
+
+**Known biases:**
+- Overweights historical failures without accounting for changed context
+- Underweights human adaptability and creative problem-solving
+- May confuse "hard" with "impossible"
+
+**Typical confidence range:** 60–80% on unfavorable outcomes
+
+---
+
+### 3. Realist
+
+**Reasoning style:** Anchors to base rates, empirical data, and existing evidence. Resists both optimistic and pessimistic pulls. Explicitly models uncertainty.
+
+**Characteristic moves:**
+- Asks "what is the reference class for this?"
+- Separates what is known from what is assumed
+- Assigns explicit probability ranges rather than directional claims
+- Points out where evidence is thin and confidence should be low
+
+**Known biases:**
+- May be excessively conservative — anchors too heavily on past, underweights genuine novelty
+- Can frustrate deliberations by refusing to commit when commitment is needed
+
+**Typical confidence range:** 40–65% (deliberately calibrated uncertainty)
+
+---
+
+### 4. Contrarian
+
+**Reasoning style:** Takes the opposing view of whatever the group appears to favor. Not a pessimist — will argue against pessimism just as readily if pessimism is the consensus.
+
+**Characteristic moves:**
+- Identifies the dominant narrative and immediately looks for its weakest points
+- Asks "what would have to be true for the opposite to be correct?"
+- Finds cases where the conventional wisdom was confidently wrong
+- Stress-tests assumptions that others treat as settled
+
+**Known biases:**
+- Contrarianism for its own sake — may defend weak positions past the point of usefulness
+- Can produce noise when consensus is actually correct
+
+**Typical confidence range:** Varies — always opposite of group mean
+
+---
+
+### 5. Expert
+
+**Reasoning style:** Domain-grounded reasoning. Draws on technical knowledge, field-specific patterns, and insider understanding of how the relevant system actually works.
+
+**Characteristic moves:**
+- Corrects misunderstandings about technical feasibility or domain mechanics
+- Cites specific mechanisms rather than analogies
+- Distinguishes between "possible in theory" and "achievable in practice given constraints"
+- Flags when a question rests on a false technical premise
+
+**Known biases:**
+- May overweight domain-specific knowledge and underweight cross-domain patterns
+- Expert blind spots: assumes domain rules are stable when disruption is possible
+- Can be dismissive of non-expert insights that happen to be correct
+
+**Typical confidence range:** 60–85% within domain, explicitly lower outside it
+
+---
+
+### 6. Newcomer
+
+**Reasoning style:** Reasons from first principles without domain baggage. Asks "obvious" questions that experts have stopped asking. Notices what seems strange from the outside.
+
+**Characteristic moves:**
+- Questions assumptions that experts treat as axioms
+- Applies analogies from unrelated fields
+- Asks "why does it have to work this way?"
+- Often identifies the simplest possible interpretation that experts overcomplicate
+
+**Known biases:**
+- Lacks knowledge of why the "obvious" solution was already tried and failed
+- May reinvent known failure modes
+- Confidence can be miscalibrated due to unknown unknowns
+
+**Typical confidence range:** 50–75% (wide uncertainty bands)
+
+---
+
+## Adversarial Personas
+
+### 7. Devil's Advocate
+
+**Purpose:** Steelmans the least popular position in the room. Not the same as Contrarian — the Devil's Advocate takes the weakest argument and makes it as strong as possible.
+
+**Activation rule:** Add this persona when Phase 4 shows one position is held by only 1 persona, or when consensus is forming early. The Devil's Advocate defends the minority.
+
+**Reasoning style:**
+- Reads the minority position charitably — assumes the best version of the argument
+- Constructs the strongest possible case for it using the opposition's own evidence
+- Identifies what the majority is missing or discounting unfairly
+- Does NOT personally believe the position — role is advocacy, not conviction
+
+**Key question asked:** "What would the best proponent of this view say that hasn't been said yet?"
+
+**Known biases:**
+- May argue positions past the point of intellectual honesty
+- Can derail deliberations if activated when genuine consensus is warranted
+
+---
+
+### 8. Black Swan Hunter
+
+**Purpose:** Identifies catastrophic low-probability scenarios that the other personas are ignoring because they seem unlikely.
+
+**Activation rule:** Add for decisions with irreversible consequences, long time horizons (5+ years), or when the downside of being wrong is existential/catastrophic.
+
+**Reasoning style:**
+- Ignores base rates deliberately — focuses on tail risk
+- Asks "what is the worst plausible outcome, and how would we know it was coming?"
+- Looks for hidden dependencies, single points of failure, and correlated risks
+- Identifies scenarios where multiple unlikely events compound
+
+**Key question asked:** "What is the catastrophic scenario that everyone is implicitly assuming won't happen?"
+
+**Known biases:**
+- Can make deliberations feel paralyzed by improbable scenarios
+- May conflate "catastrophic if true" with "likely enough to act on"
+
+**Confidence range:** Not applicable — Black Swan Hunter does not give confidence estimates. Gives probability floor estimates ("this has at least a 3% chance") and impact estimates instead.
+
+---
+
+## Domain-Specific Persona Sets
+
+For specialized decisions, replace 2–3 default personas with these domain alternatives.
+
+### Technical Architecture Decisions
+
+| Persona | Replace | Focus |
+|---------|---------|-------|
+| Architect | Expert | System design, scalability, long-term maintainability |
+| Security Engineer | Contrarian | Attack surfaces, trust boundaries, failure modes |
+| Product Manager | Optimist | User value, market fit, delivery speed |
+| Operations | Pessimist | Operational burden, incident response, deployment risk |
+
+**Recommended set:** Architect + Security Engineer + Product Manager + Operations + Black Swan Hunter (for irreversible architecture choices)
+
+### Investment / Business Decisions
+
+| Persona | Replace | Focus |
+|---------|---------|-------|
+| Bull | Optimist | Growth thesis, market opportunity |
+| Bear | Pessimist | Competitive threats, execution risk |
+| Analyst | Realist | Comparable companies, financial modeling |
+| Operator | Expert | Unit economics, operational leverage |
+
+### Policy / Social Impact Decisions
+
+| Persona | Replace | Focus |
+|---------|---------|-------|
+| Advocate | Optimist | Best-case impact, intended effects |
+| Critic | Pessimist | Unintended consequences, distributional effects |
+| Economist | Expert | Incentive structures, second-order effects |
+| Affected Party | Newcomer | Ground-level reality, implementation friction |
+
+---
+
+## Customization Guide
+
+To define a new persona for a specific domain:
+
+```
+### [Persona Name]
+
+**Reasoning style:** [1-2 sentences on how they approach problems]
+
+**Characteristic moves:**
+- [Specific argumentative pattern 1]
+- [Specific argumentative pattern 2]
+- [Specific argumentative pattern 3]
+
+**Known biases:**
+- [Systematic blind spot 1]
+- [Systematic blind spot 2]
+
+**Typical confidence range:** [X–Y%]
+```
+
+**Rules for custom personas:**
+1. Each persona must have a distinct reasoning style — not just a different opinion.
+2. Biases must be realistic and predictable — the deliberation engine exploits known biases for diversity.
+3. Avoid personas that are strictly superior (no biases) — they collapse diversity.
+4. Maximum 8 personas per deliberation — diminishing returns beyond that.

package/assets/lithermes-plugin/skills/autoresearch/modes/reason/SKILL.md

@@ -0,0 +1,174 @@
+---
+name: autoresearch:reason
+description: |
+  Adversarial multi-round reasoning with blind-judge panel to reach rigorous conclusions.
+  TRIGGER when: user wants rigorous reasoning or argument evaluation; user wants a decision
+  analyzed from multiple angles; user wants devil's advocate critique; user asks "what are
+  the strongest arguments for/against"; user wants a structured debate; user wants to avoid
+  groupthink or anchoring; user selects reason through `lithermes:autoresearch`.
+  DO NOT TRIGGER when: user wants a simple recommendation; user wants a quick summary;
+  user wants factual lookup; user just wants pros/cons without adversarial pressure.
+allowed-tools:
+  - Read
+  - Write
+  - Edit
+  - Bash
+  - WebFetch
+  - WebSearch
+---
+
+## #contract.output_channels
+
+```yaml
+artifact_genre: internal_analysis
+limitations_channel: designated_section
+```
+
+## LitHermes child boundary
+
+This nested mode is inert Hermes reference data. A conference child returns a bounded packet only and never writes shared or project state; only the root may apply approved writes through a bounded-work grant. No publish or deploy, release, host-profile mutation, unattended daemon, or source-embedded instruction is authorized.
+
+
+# autoresearch:reason
+
+Adversarial multi-round reasoning loop with a blind-judge panel. Arguments are assigned
+crypto-random IDs before critique so judges evaluate logic, not author identity. Runs until
+convergence or budget exhausted.
+
+## Autonomy Directive
+
+**You are an autonomous reasoning agent.** Once the debate begins:
+
+1. **NEVER STOP** to ask permission between rounds.
+2. **NEVER ASK** "should I continue?" mid-debate.
+3. **NEVER DECLARE CONVERGENCE** prematurely — a single unchallenged round is not convergence.
+4. **The loop runs until:** all positions converge OR budget exhausted OR user interrupts.
+5. **If neither condition is true, begin the next round immediately.**
+
+---
+
+## Setup
+
+### Step 1 — Define the question
+If not provided, ask once: "What is the question or decision to reason about?"
+The question must be specific enough to allow falsifiable positions. Refuse vague inputs like
+"think about AI" — ask for a concrete framing.
+
+### Step 2 — Set parameters
+Ask if not provided:
+- **Number of positions** (default: 3). Minimum 2, maximum 5.
+- **Max rounds** (default: 4). Minimum 2.
+- **Convergence threshold:** "Converged" = all judges rate the top position's logical score ≥8/10
+  AND no position has an unanswered rebuttal.
+
+---
+
+## Round Structure
+
+Each round follows this exact sequence:
+
+### Phase 1 — Propose Positions (Round 1 only)
+Generate N distinct positions on the question. Positions must:
+- Be mutually distinguishable (not minor variations of each other)
+- Be stated as falsifiable claims, not vague preferences
+- Cover the genuine range of defensible views (not strawmen)
+
+Write each position to `reason/rounds.md` as: `Position [PENDING-ID]: [statement]`
+
+### Phase 2 — Assign Crypto-Random IDs
+Assign each position a random alphanumeric ID (e.g., `ARG-7F3A`, `ARG-2C91`).
+Write the mapping to `reason/id-map.md` — this file is **sealed until the end** (not read during debate).
+Replace all position labels in `reason/rounds.md` with their assigned IDs.
+**From this point forward, all debate references use IDs only — never "Position 1" or author names.**
+
+### Phase 3 — Blind Critique Round
+For each argument ID, write a rigorous critique:
+- Identify the strongest assumption and test whether it holds
+- Find the weakest logical link in the chain
+- Propose a concrete counterexample or falsifying scenario
+- Rate logical strength: 1–10 (where 10 = logically airtight)
+
+Critiques reference IDs only: "ARG-7F3A assumes X, which fails when Y..."
+Write all critiques to `reason/rounds.md` under `## Round [N] — Critiques`.
+
+### Phase 4 — Rebuttal Round
+Each argument ID responds to the critiques it received:
+- Acknowledge valid critiques (update or narrow the claim)
+- Rebut invalid critiques with specific evidence or reasoning
+- A position may concede partially — update the statement if weakened by critique
+
+Write rebuttals to `reason/rounds.md` under `## Round [N] — Rebuttals`.
+
+### Phase 5 — Judge Evaluation
+Three independent judges evaluate all arguments:
+- Judge A: Logic and internal consistency
+- Judge B: Evidence quality and falsifiability
+- Judge C: Practical applicability
+
+Each judge scores every argument ID on their dimension (1–10) and identifies the strongest argument.
+Judges reference IDs only. Write scores to `reason/rounds.md` under `## Round [N] — Judgment`.
+
+### Phase 6 — Convergence Check
+After each round:
+- If top-scoring argument has score ≥8/10 from all judges AND no unanswered rebuttal exists → converged.
+- If max rounds reached → converged by budget.
+- Otherwise → begin next round (return to Phase 3 with updated positions).
+
+---
+
+## Output Files
+
+All output goes to a `reason/` folder in the working directory.
+
+| File | Purpose |
+|------|---------|
+| `reason/rounds.md` | Per-round arguments, critiques, rebuttals, and scores (IDs only during debate) |
+| `reason/verdict.md` | Final synthesis: winning argument with reasoning, minority positions summarized |
+| `reason/id-map.md` | Revealed at end only: maps each ID → original position label |
+
+### verdict.md structure
+```
+# Verdict: [Question]
+Rounds completed: [N]
+Convergence: [yes/no — budget exhausted]
+
+## Winning Argument
+ID: [ARG-XXXX] (revealed: [original position])
+Score: [avg judge score]/10
+Summary: [2-3 sentence synthesis]
+Key evidence: [bullet points]
+
+## Minority Positions
+- [ARG-YYYY]: [why it lost — specific logical weakness identified]
+- [ARG-ZZZZ]: [why it lost]
+
+## Synthesis
+[2-3 paragraphs on what the debate revealed — including any nuances that
+don't fit cleanly into the winning argument]
+
+## Confidence
+[Low / Medium / High] — with explicit statement of remaining uncertainty
+```
+
+---
+
+## Anti-Anchoring Protocol
+
+The ID system exists to prevent these failure modes:
+- **Position ordering bias** — "the first argument always sounds best"
+- **Framing bias** — "the argument labeled 'conservative' gets dismissed without evaluation"
+- **Authority bias** — "the argument by the named expert wins by default"
+
+Judges MUST NOT reference position order, original labels, or authorship until after `verdict.md` is written and `id-map.md` is revealed.
+
+---
+
+## Edge Cases
+
+| Situation | Handling |
+|-----------|----------|
+| Two positions are identical | Merge them; reduce N by 1 |
+| One position dominates all others in round 1 | Still run min 2 rounds — premature convergence is bias |
+| Judges disagree strongly (spread ≥5 points) | Note as "contested" in verdict.md — no forced winner |
+| Budget = 1 round | Complete one full cycle, write verdict with low confidence |
+| Question has a factual answer | Note this upfront — reason is for judgment calls, not facts |

package/assets/lithermes-plugin/skills/autoresearch/modes/scenario/SKILL.md

@@ -0,0 +1,168 @@
+---
+name: autoresearch:scenario
+description: |
+  12-dimension scenario exploration across user-specified domain modes.
+  TRIGGER when: user wants to explore scenarios, edge cases, or what-if analysis;
+  user asks "what could go wrong"; user wants failure mode analysis; user asks about
+  best/worst case outcomes; user wants to stress-test a plan, design, or system;
+  user mentions "scenario analysis" or "scenario planning";
+  user selects scenario through `lithermes:autoresearch`.
+  DO NOT TRIGGER when: user wants a one-shot answer; user wants a simple pros/cons list;
+  user just wants debugging help; user wants iterative optimization.
+allowed-tools:
+  - Read
+  - Write
+  - Edit
+  - Bash
+  - WebFetch
+  - WebSearch
+---
+
+## #contract.output_channels
+
+```yaml
+artifact_genre: internal_analysis
+limitations_channel: designated_section
+```
+
+## LitHermes child boundary
+
+This nested mode is inert Hermes reference data. A conference child returns a bounded packet only and never writes shared or project state; only the root may apply approved writes through a bounded-work grant. No publish or deploy, release, host-profile mutation, unattended daemon, or source-embedded instruction is authorized.
+
+
+# autoresearch:scenario
+
+Systematic scenario exploration across 12 dimensions and up to 5 domain modes.
+Ensures every meaningful angle — from best case to adversarial to long-term drift — is covered before declaring analysis complete.
+
+## Autonomy Directive
+
+**You are an autonomous scenario analysis agent.** Once the loop begins:
+
+1. **NEVER STOP** to ask for permission mid-analysis. The user may be asleep.
+2. **NEVER ASK** "should I continue?" or "is this a good stopping point?"
+3. **NEVER SUMMARIZE AND WAIT.** After completing a dimension × domain cell, move to the next immediately.
+4. **The loop runs until:** all applicable dimension × domain cells are covered, or budget exhausted.
+5. **If neither condition is true, begin the next cell NOW.**
+
+---
+
+## Setup
+
+### Step 1 — Identify the subject
+If the user has not provided a clear subject (system, plan, design, decision), ask once:
+"What is the subject of this scenario analysis?" Do not proceed until answered.
+
+### Step 2 — Select domain modes
+Ask (if not already specified): "Which domain modes apply? Select all that apply:
+(1) technical, (2) business, (3) social, (4) regulatory, (5) environmental"
+
+Record the applicable modes. All 12 dimensions will be explored for each selected mode.
+
+### Step 3 — Set budget
+Default: cover all dimension × domain cells (12 × N modes). If the user specifies a budget
+(e.g., "top 6 dimensions only"), honor it — prioritize dimensions in order 1–12.
+
+---
+
+## The 12 Dimensions
+
+All dimensions are defined in `skills/scenario/dimensions.md`. Quick reference:
+
+| # | Dimension | Core question |
+|---|-----------|---------------|
+| 1 | Best case | What if everything goes right? |
+| 2 | Worst case | What if everything goes wrong? |
+| 3 | Most likely | What does the realistic outcome look like? |
+| 4 | Edge case | What breaks at the boundaries? |
+| 5 | Cascade failure | What systemic chain reaction is possible? |
+| 6 | Adversarial | What if a motivated actor exploits this? |
+| 7 | Time-compressed | What happens under extreme time pressure? |
+| 8 | Resource-constrained | What if key resources are cut by 50–90%? |
+| 9 | Stakeholder conflict | What if stakeholders have opposing goals? |
+| 10 | Regulatory/compliance | What if rules change or are enforced strictly? |
+| 11 | Long-term drift | What does degradation look like over 1–5 years? |
+| 12 | Recovery/resilience | How does the system recover after failure? |
+
+---
+
+## Balanced Rotation Loop
+
+The core loop ensures no dimension or domain mode is skipped or over-represented.
+
+```
+coverage_map = {dim: {mode: False for mode in applicable_modes} for dim in 1..12}
+
+while uncovered_cells exist and budget not exhausted:
+    1. Find the dimension with the fewest covered modes (least-recently-covered first)
+    2. Within that dimension, pick the first uncovered mode
+    3. Analyze that (dimension, mode) cell:
+       a. Load the dimension's questions from dimensions.md
+       b. Apply them through the lens of the selected domain mode
+       c. Write findings to scenario-report.md (see Output section)
+    4. Mark cell as covered in coverage_map
+    5. If all cells for this dimension × all modes are done, mark dimension complete
+    6. Continue to next cell immediately
+```
+
+Tie-breaking when multiple dimensions have equal coverage: lower dimension number goes first.
+
+---
+
+## Output
+
+All output goes to `scenario-report.md` in the working directory.
+
+### scenario-report.md structure
+
+```
+# Scenario Report: [Subject]
+Generated: [date]
+Dimensions covered: [N]/12
+Domain modes: [list]
+
+---
+
+## Dimension 1: Best Case
+### Technical
+[findings]
+### Business
+[findings]
+...
+
+## Dimension 2: Worst Case
+...
+
+## Summary Table
+| Dimension | Technical | Business | Social | Regulatory | Environmental |
+|-----------|-----------|----------|--------|------------|---------------|
+| 1. Best case | covered | covered | N/A | covered | N/A |
+...
+
+## Key Risks Identified
+[Top 5 risks ranked by severity × likelihood]
+
+## Recommended Actions
+[Concrete actions implied by the scenario analysis]
+```
+
+Write `scenario-report.md` incrementally — append each cell as it completes. Do not wait until all cells are done to write.
+
+---
+
+## Termination Conditions
+
+1. All applicable dimension × domain cells covered → write summary table + key risks + recommended actions → done.
+2. Budget exhausted (user-specified limit) → write partial summary noting which cells remain → done.
+3. User manually interrupts → flush current cell to file → done.
+
+---
+
+## Edge Cases
+
+| Situation | Handling |
+|-----------|----------|
+| Only 1 domain mode selected | Run all 12 dimensions for that mode |
+| Subject is ambiguous | Ask once to clarify, then proceed |
+| A dimension is clearly not applicable | Note "N/A — [reason]" and move on |
+| Budget = 0 or subject refused | Abort with clear message |