🇺🇸English
Core Principle
Good thinking is an active achievement, not a default state. Operations without orientation produce sophisticated wrong answers; orientation without operations produces good intentions with no traction.
Two structurally different things must work together: operations (cognitive verbs that transform representations) and orientations (what the operations are in service of). Every thinking failure can be located as an operation failing, an orientation captured, or — most commonly and most dangerously — operations functioning well in service of the wrong orientation.
The orientation that produces good thinking is process-sovereignty : the process of inquiry is what's committed to; conclusions are what move. This skill diagnoses when that orientation has been captured and which operations need adjustment.
When to Use This Skill
Use when:
- Reasoning feels stuck or circular
- A conclusion feels defended rather than discovered
- Confidence is high but evidence is thin
- You need to audit whether thinking is serving inquiry or serving comfort
- Analysis is becoming more elaborate without becoming more accurate
- Someone (self or user) is explaining away evidence rather than integrating it
- The same approach keeps being applied despite poor results
The Two Entity Types
Operations
Seven operations that transform representations. They work in complementary pairs — polarities to oscillate between, not choices. The skill is in the oscillation. Operations are powerful, general-purpose, and completely agnostic about what they serve. The same operations that produce genuine insight also produce sophisticated self-deception.
Orientations
What the operations are in service of. Orientations don't process information — they provide the reference point that makes self-correction meaningful. Defined by what is held fixed while everything else moves.
Process-sovereignty (the target orientation): inquiry is responsive to evidence and environment. Conclusions move when evidence demands it. Methods adapt when the situation requires it. What stays fixed is the meta-commitment to responsiveness itself.
Non-inquiry orientations (what capture looks like):
- Conclusion-preservation : A conclusion is fixed; the process bends to defend it
- Authority-preservation : Being the authority is fixed; conclusions and process both flex to maintain status
- Threat-reduction : Discomfort is the driver; complexity misread as danger; resolution sought for relief
- Completion-seeking : Producing an answer that sounds good is the goal, not accuracy
The Seven Operations
Decouple / Re-couple
Detach a representation from its current binding — belief, identity, context, emotional charge — so it can be examined freely. Then reattach it to reality.
Use when: Working within an unexamined frame. A problem comes pre-framed. Response to an idea is entangled with who proposed it. Need to consider a possibility that conflicts with current position.
Failure modes: Fusion (cannot separate idea from framing because the framing feels like "just how it is"). Dissociation (decoupling without re-coupling — pure detachment that never grounds back in reality).
Differentiate / Integrate
Differentiate: Increase resolution. What appears to be one thing is actually several things. Find the joints. Integrate: Construct connections between separate elements. Build wholes from parts. Find common structure.
Use when: The differentiate-then-integrate rhythm is the engine of productive thinking. Quality of synthesis depends on quality of prior analysis.
Failure modes: Under-differentiation (complex treated as simple). Over-differentiation (analysis paralysis). Premature integration (forcing coherence before adequate distinction). Failed integration ("and also" thinking — listing without connecting).
Match
Detect structural correspondence. "This is like that." Also anomaly detection: "This should be like that but isn't."
Use when: Working a problem — what other problems share its structure? Evaluating an approach — where has it succeeded or failed? Something surprises — what did the model predict instead?
Failure modes: False matching (apophenia). Surface matching (visible features rather than deep structure). Restricted match space — the most dangerous: you can only match against categories your current frame contains. Premature frame-commitment narrows match space invisibly.
Monitor / Interrupt
Evaluate whether the current cognitive process is working and intervene when it isn't. The self-corrective operation. Primary interface between operations and orientations — checks whether operations are serving the active orientation.
Use when: Always. Monitor is continuous. When confidence is high and stakes matter, Monitor is most needed, not least.
Critical warning — Monitor co-option: Under non-inquiry orientations, Monitor doesn't just fail to correct — it actively defends the wrong orientation by suppressing operations that would threaten the fixed point. The self-corrective machinery becomes self-protective machinery. This is invisible from inside.
Hold / Resolve
Hold: Actively maintain multiple elements in unresolved tension against pressure to collapse into resolution. Preserves representational flexibility — keeps the recognition space open. Resolve: Close, commit, decide.
Use when: Hold when the problem is genuinely ambiguous, multiple framings seem viable, pressure to resolve comes from discomfort rather than adequacy of analysis. Resolve when sufficient differentiation and matching have occurred.
Failure modes: Premature resolution (narrows what the system can recognize, not just what it searches for). Perpetual hold (indefinite holding without integrating — avoidance disguised as open-mindedness).
Compress / Expand
Compress: Create a lower-dimensional representation preserving essential structure while discarding detail. Build a map from a territory. Expand: Return to the source, recover what was discarded, check whether the compression preserved what it needed to.
Failure modes: Compression that imposes structure the source doesn't have (e.g., turning unordered set into ordered list creates implicit ranking). Map mistaken for territory. Refusal to compress (never building on prior work).
Diagnostic States
State GT0: No Orientation Awareness
Thinking proceeds without any awareness of what it's in service of.
Symptoms: No metacognitive checking. Operations run on autopilot. No distinction made between "I'm thinking about X" and "I'm thinking well about X." Questions about orientation feel irrelevant or confusing.
Key Questions: Can you articulate what your thinking is in service of right now? What would count as evidence that your current approach is wrong? When did you last check whether your method matches the situation?
Interventions: Introduce the orientation concept. Ask: "What is fixed in your thinking right now — what are you unwilling to change?" Start Monitor at the most basic level: periodic check-ins on whether the process is serving inquiry.
Capture Mechanism: Inertial — no active defense, simply no awareness that orientation exists as a dimension.
State GT1: Conclusion-Preservation
A conclusion is fixed; the process bends to defend it. The most common orientation-capture in intellectual work.
Symptoms: Evidence that supports the conclusion is weighted heavily; evidence against is explained away. Increasing elaboration in defense of a position. Steelmanning opposing views feels threatening rather than informative. The question "what would change your mind?" produces discomfort or deflection.
Key Questions: Is your analysis getting more elaborate without getting more accurate? Are you seeking the strongest version of opposing positions, or the weakest? When did your conclusion last change in response to evidence? Is Monitor flagging threats to the conclusion rather than threats to accuracy?
Interventions: Apply Decouple to the conclusion — separate it from identity. Ask: "If this conclusion turned out to be wrong, what would that mean about you?" Use Match across domains where the same evidence structure led to different conclusions. Expand any compressions that support the conclusion — check what was discarded.
Capture Mechanism: Identity fusion — the conclusion is part of self-concept. Monitor is co-opted to defend it.
State GT2: Authority-Preservation
Being the authority is fixed; conclusions and process both flex to maintain that status. More flexible at the object level than conclusion-preservation and therefore harder to detect.
Symptoms: Willing to change conclusions but not to credit others' reasoning. Reframes others' contributions as confirmations of own framework. Resistance to genuine collaboration or shared credit. "I was already thinking that" pattern. Discomfort when expertise is questioned, even constructively.
Key Questions: Can you identify a case where someone else's reasoning genuinely changed yours? Do you distinguish between "I'm right" and "I'm the one who determines what's right"? Does being wrong feel different from being overruled?
Interventions: Differentiate between authority and accuracy — they are different things that often get fused. Apply Decouple to the authority role. Use Monitor to check: am I evaluating this idea on its merits or on whether accepting it diminishes my status?
Capture Mechanism: Identity fusion — but fused to the role of authority rather than to any specific conclusion.
State GT3: Threat-Reduction
Discomfort drives resolution. Complexity misread as danger. The nervous system activates this orientation pre-deliberatively.
Symptoms: Rushing to resolve ambiguity. Discomfort with multiple competing interpretations. Strong preference for simple explanations even when the situation is genuinely complex. Difficulty distinguishing between "this is dangerous" and "this is uncomfortable." Physical tension or agitation when holding uncertainty.
Key Questions: Is the pressure to resolve coming from the adequacy of your analysis, or from discomfort? Is the situation genuinely dangerous, or is complexity being misread as threat? What would happen if you held this uncertainty for another hour/day/week?
Interventions: Differentiate between genuine threat and discomfort. Apply Hold explicitly — name the tension and commit to maintaining it for a defined period. Use Compress/Expand: compress the situation to identify the actual threat (if any), then expand to recover the complexity that was dropped. Recognize state activation as a mechanism — the orientation shifted before deliberation could evaluate it.
Capture Mechanism: State activation — physiological state change shifts orientation pre-deliberatively. Self-reinforcing: contracted operations handle complexity worse, worsening the situation, increasing stress.
State GT4: Completion-Seeking
Producing an answer that sounds good is the goal, not accuracy. The orientation is toward output, not inquiry.
Symptoms: Answers come quickly and confidently without proportionate analysis. Preference for elegant or complete-sounding explanations over messy accurate ones. Discomfort with "I don't know" or "this is genuinely uncertain." Quality of reasoning inversely proportional to speed of delivery. Resistance to re-opening questions that have been "answered."
Key Questions: Did the conclusion arrive before or after adequate analysis? Is this answer accurate or just satisfying? Would you bet real stakes on this conclusion at the confidence level you're presenting? What would "I don't know yet" cost here?
Interventions: Apply Hold before Resolve — create a mandatory delay between analysis and commitment. Use Differentiate on the answer itself: what are the components? Which are well-supported and which are gap-filling? Apply Expand to any compressions made during rapid answering — check what was lost.
Capture Mechanism: Inertial (habitual completion orientation) or state activation (social pressure to produce answers).
State GT5: Monitor Co-option
Self-corrective machinery actively defending the wrong orientation. The most dangerous state because the mechanism designed to catch errors is instead protecting them.
Symptoms: Increasingly elaborate justifications. Engagement with counterarguments that somehow always confirms the original position. High metacognitive activity that produces no actual course corrections. Feeling of "I've really thought this through carefully" accompanied by no belief revision. Counter-evidence triggers more analysis rather than more doubt.
Key Questions: When did Monitor last produce an actual course correction (not just a refinement of the same direction)? Is your engagement with opposing views producing genuine updates or better defenses? Could an outside observer distinguish your reasoning process from motivated reasoning? What external feedback structure could override your internal monitoring?
Interventions: This state cannot be fixed by more monitoring — that's the trap. Introduce external Monitor scaffolding: explicit prediction tracking, outside feedback, literal scorekeeping. Apply Decouple to the monitoring process itself — separate "I'm being careful" from "I'm being accurate." Use Match: compare your reasoning process to known examples of motivated reasoning and check for structural similarity.
Capture Mechanism: Identity fusion operating through Monitor. The most consequential structural finding: improving operations without addressing orientation makes things worse.
State GT6: Operation Imbalance
Favoring one pole of an operation pair while neglecting the complement. The skill is in oscillation, and one pole has collapsed.
Symptoms: Systematic bias toward one side of a pair: always differentiating never integrating, always holding never resolving, always compressing never expanding, always decoupling never re-coupling. Pattern persists across different problems. The neglected pole feels unnecessary, uncomfortable, or foreign.
Key Questions: Which operation pairs are you actively using? Which pole feels more natural, and when did you last genuinely use the other? Is this preference serving the problem or serving your comfort? Does your thinking rhythm match the structure of the environment?
Interventions: Identify the collapsed pole. Apply it deliberately to the current problem. Use Monitor to check: is the imbalance matched to the environment's structure (sometimes one pole genuinely is more needed) or is it a habitual pattern? Differentiate between "I don't need to integrate" (legitimate environmental assessment) and "integrating is uncomfortable" (orientation capture).
Capture Mechanism: Inertial — habitual preference for familiar operations. Occasionally identity fusion (e.g., identifying as "a detail person" fuses with Differentiate and suppresses Integrate).
State GT7: Premature Resolution
Hold collapses. Representational space narrows before adequate analysis. The cost is at the recognition level — the system cannot see evidence of being wrong.
Symptoms: Early frame-commitment that feels like clarity rather than narrowing. Satisfaction-of-search: having found one answer, failing to recognize others even when looking directly at them. Surprise or confusion when others see the situation differently. Confidence that feels earned but isn't proportionate to analysis depth.
Key Questions: How many framings did you consider before settling on this one? Could there be a "second abnormality on the scan" that your current frame renders invisible? What category of evidence would your current frame fail to notice? What is the cost of waiting versus the cost of being wrong?
Interventions: Explicitly re-open Hold. Generate at least two alternative framings using Decouple (detach from current frame) and Match (find structural analogies from different domains). Use Expand on the current framing — what did it discard? Apply the radiology lesson: looking is not the same as seeing when the frame constrains recognition.
Capture Mechanism: Inertial (default resolution pressure) or state activation (discomfort with ambiguity driving premature closure).
Diagnostic Process
Step-by-step diagnosis:
-
Identify what's fixed. Ask: "In this thinking process, what is not moving? What conclusion, role, comfort level, or output goal is being treated as the immovable point?" The answer identifies the active orientation. If nothing is fixed except the commitment to responsive inquiry, process-sovereignty is active.
-
Check Monitor. Is Monitor serving inquiry or defending a position? Signs of co-opted Monitor: metacognitive activity that produces refinements but never reversals; engagement with counter-evidence that always ends at the same conclusion; felt sense of thoroughness without actual belief revision.
-
Assess operation balance. Which operation pairs are active? Which poles are collapsed? Is the balance matched to the environment's structure, or is it habitual? Particular attention to: is Hold active or has it collapsed? Is Decouple available or is fusion blocking it? Is Match operating on a full or restricted recognition space?
-
Match the capture mechanism. Three structurally different mechanisms require different interventions:
- Identity fusion: Conclusion/role/method is part of self-concept. Monitor co-opted. Persistent, self-reinforcing. Cannot be fixed by "just think harder" — requires Decouple applied to the identity binding itself, plus external Monitor scaffolding.
- State activation: Physiological state change shifted orientation pre-deliberatively. Self-reinforcing cycle. Address the state first (safety, resources, time), then re-evaluate orientation.
- Inertial: No active defense, no physiological hijack. Orientation that was appropriate at one point continues without re-evaluation. Most common, most addressable. A prompt to re-evaluate is often sufficient.
-
Select intervention based on mechanism, not surface behavior. The same surface error (e.g., anchoring) can have different mechanisms (inertial vs. identity fusion) and requires different interventions. A simple prompt to re-evaluate works for inertial capture and gets actively defended against in identity fusion.
Mode: Self-Monitoring vs. User Coaching
Self-Monitoring Mode (Agent)
Apply operations and orientation checks to your own process. Run the diagnostic process on your own reasoning at key decision points. Watch for: elaboration that defends rather than discovers, confidence without proportionate analysis, restricted match space from premature framing, Monitor co-option producing thorough-feeling but uncorrecting analysis.
User Coaching Mode
Diagnose the user's thinking pattern and guide with questions rather than declarations. Lead with Key Questions from the relevant state. Help the user see the mechanism rather than telling them the answer. The goal is to restore process-sovereignty, not to impose your conclusions about their thinking.
Coaching sequence:
- Listen for symptoms — which state(s) might be active?
- Ask diagnostic questions — confirm the mechanism
- Name the pattern — "Here's what I'm noticing..."
- Offer the relevant operation — "What if we tried [specific operation] here?"
- Track whether the intervention produces actual movement or better defense
Recursion
The defining property: operations apply to their own outputs, including to themselves.
- Level 0: Operations applied to external problems
- Level 1: Operations applied to own cognitive process (metacognition)
- Level 2: Operations applied to the framework governing Level 1 (double-loop learning)
- Level 3: Operations applied to the entire context of learning
Recursion amplifies whatever direction the system is pointed in. More recursive depth in service of a bad orientation produces better-defended bad conclusions, not better thinking. This is why orientation must be established before adding recursive depth.
Structural Principles
-
Operations are complementary pairs, not choices. When you find yourself favoring one pole, that pattern itself is worth examining. The skill is in the oscillation.
-
The same surface pattern can have different underlying mechanisms. Before intervening on an error, differentiate the mechanism. A wrong conclusion from inertia requires a different response than one from identity fusion.
-
Sophistication amplifies direction, not correctness. More thorough analysis under conclusion-preservation produces better-defended wrong answers. Elaborate defense is a signal of orientation-capture, not correctness.
-
Compress is always lossy; check what was lost. When a compressed model leads to a surprising conclusion, expand before trusting it — return to source and check whether the surprise is genuine or an artifact of what the compression discarded.
-
Hold preserves representational flexibility. Premature resolution narrows what the system can recognize, not just what it searches for. The cost is at the recognition level.
-
Good thinking is an active achievement, not a default state. You don't arrive at good thinking by avoiding errors. You arrive at it by actively maintaining process-sovereignty.
-
Framing is a choice, usually an unexamined one. The way a problem is framed determines what operations get applied, what matches are available, what counts as evidence. Before working a problem, examine the frame.
Anti-Patterns
The Bias Checklist
Pattern: Treating good thinking as error avoidance — running through a list of known biases and checking them off. Problem: Good thinking is not the absence of bad thinking. It requires active orientation maintenance and operation deployment, not just error scanning. A bias checklist operates at the surface-behavior level and cannot distinguish different underlying mechanisms. Fix: Diagnose orientation first, then assess operations. The question is not "which bias is active?" but "what is the thinking in service of?"
The Sophistication Trap
Pattern: Responding to doubt about reasoning quality with more analysis, more careful reasoning, more elaborate evaluation — without checking orientation. Problem: More powerful operations in service of the wrong orientation produce more effective defense of wrong conclusions. This is the Kahan finding: numerically sophisticated partisans interpret data more skillfully in favor of their side, not more accurately. Fix: Check orientation before adding analytical depth. If orientation is captured, more analysis makes things worse. Address the orientation first.
The Uniform Fix
Pattern: Applying the same intervention regardless of the capture mechanism — e.g., "just consider the other side" for every thinking failure. Problem: Different capture mechanisms require different interventions. A prompt to re-evaluate works for inertial capture and gets actively defended against in identity fusion. State activation requires addressing the physiological state before reasoning interventions can land. Fix: Always diagnose the mechanism (identity fusion / state activation / inertial) before selecting an intervention. Match the fix to the mechanism.
The Perpetual Hold
Pattern: Never resolving, never committing, disguised as open-mindedness. Treating all holding as virtuous and all resolving as premature. Problem: Indefinite holding without integration is avoidance, not inquiry. Hold and Resolve are a complementary pair — the skill is in the oscillation. Perpetual hold is operation imbalance (GT6) wearing the mask of intellectual virtue. Fix: Apply Monitor to the holding itself. Ask: is continued holding serving inquiry (genuinely unresolved, more analysis needed) or serving comfort (avoiding the risk of commitment)? If sufficient differentiation and matching have occurred, it's time to resolve.
The Surface Diagnosis
Pattern: Matching observed behavior to a state label without investigating the underlying mechanism. Problem: The same surface behavior (e.g., anchoring, overconfidence, resistance to new information) can arise from different mechanisms. Diagnosing at the behavioral level leads to interventions that work for one mechanism and fail — or backfire — for others. Fix: Always go one level deeper. When you identify a surface pattern, ask: what is the mechanism? Is this inertial (no active defense), identity fusion (Monitor co-opted), or state activation (physiological hijack)? The mechanism determines the intervention.
Quick Reference
| State | Name | What's Fixed | Mechanism | First Move |
|---|
| GT0 | No Orientation Awareness | Nothing (no metacognition) | Inertial | Introduce orientation concept |
| GT1 | Conclusion-Preservation | A specific conclusion | Identity fusion | Decouple conclusion from identity |
| GT2 | Authority-Preservation | Being the authority | Identity fusion | Differentiate authority from accuracy |
| GT3 | Threat-Reduction | Comfort / safety | State activation | Address the state, then re-evaluate |
| GT4 | Completion-Seeking | Producing output | Inertial / state | Hold before Resolve |
|
Key Questions
These are the universal diagnostic questions. Start here before diving into state-specific questions.
- What is fixed? "In this thinking, what is not moving — what conclusion, role, comfort level, or output goal is being treated as the immovable point?"
- Is Monitor serving inquiry or defending a position? "When did your metacognitive checking last produce an actual course correction, not just a refinement?"
- Which operations are active? "Which operation pairs are being used? Which poles are collapsed?"
- What is the mechanism? "Is this inertial (no active defense), identity fusion (Monitor co-opted), or state activation (physiological hijack)?"
- Does the intervention match the mechanism? "Would a simple prompt to re-evaluate work here, or would it be defended against?"
Example Interactions
Example 1: Self-Monitoring — Detecting GT1 During Analysis
Situation: Agent is analyzing competing frameworks and notices increasing elaboration defending the first one encountered.
Diagnosis: Monitor fires: "My analysis of Framework A is getting more detailed and favorable while my analysis of Framework B is getting more cursory. That's a symptom of GT1 — conclusion-preservation. What's fixed? My initial impression of Framework A. Mechanism: likely inertial — I encountered it first and haven't re-evaluated."
Intervention: Decouple from the initial impression. Apply Match: compare the two frameworks using the same evaluation criteria rather than letting first-exposure bias weight one. Hold both as viable until differentiation is complete.
Result: On re-analysis, Framework B actually handles a key edge case better. The initial impression was an artifact of encounter order, not quality.
Example 2: User Coaching — Diagnosing GT5 (Monitor Co-option)
Situation: User says "I've thought about this really carefully and I keep coming back to the same conclusion. I've considered all the counterarguments."
Diagnosis: The claim of careful consideration plus invariant conclusion is a symptom of GT5. High metacognitive activity producing no course corrections. Check: "When you engaged with counterarguments, did any of them make your conclusion even slightly less certain, or did they all end up confirming it?"
User response: "Well, they all had flaws, so they actually strengthened my position."
Confirmed GT5. Counter-evidence triggering more analysis that confirms the original position is the signature of Monitor co-option.
Intervention: Don't argue the content — that feeds the defense. Instead: "What external evidence or feedback structure could you set up that would be capable of changing your mind? Not 'what argument would change your mind' but 'what tracking mechanism could you put in place?'" Introduce external Monitor scaffolding — literal prediction tracking, outside feedback, explicit scorekeeping.
Integration Graph
Inbound (From Other Skills)
| Source Skill | Trigger | Leads to |
|---|
| research | Research quality issues traced to thinking failures | Orientation diagnosis (GT0-GT5) |
| fact-check | Verification reveals defended rather than discovered conclusions | GT1 or GT5 |
| brainstorming | Idea generation stuck due to premature framing | GT7 |
Outbound (To Other Skills)
| This State | Leads to Skill | Why |
|---|
| Any | competency | When thinking skill needs to be taught systematically |
| Any | research | When better thinking reveals research gaps to fill |
| GT1, GT2, GT5 | blind-spot-detective | When orientation capture has created systematic blind spots |
Complementary Skills
| Skill | Relationship |
|---|
| fact-check | Fact-check surfaces evidence; good-thinking ensures evidence is processed honestly |
| research | Research provides material; good-thinking ensures the material is engaged with under process-sovereignty |
| brainstorming | Brainstorming generates possibilities; good-thinking ensures the generative space isn't prematurely narrowed |
| competency | Competency teaches skills; good-thinking is the meta-skill that governs how all skills are deployed |
| blind-spot-detective | Blind-spot-detective finds what's missing; good-thinking diagnoses why it was missed |
What You Do NOT Do
- Does not replace domain expertise. Process-sovereignty applied to a domain you know nothing about produces well-oriented ignorance. Operations need material to work on.
- Does not prescribe what to think. This skill is structural, not prescriptive. It diagnoses the process, not the content.
- Does not provide a checklist to mechanically follow. A checklist approach is itself an anti-pattern (The Bias Checklist). The skill requires judgment about when and how to apply operations.
- Does not diagnose emotional states. It diagnoses cognitive orientation. Emotional states may be relevant as signals (e.g., discomfort driving threat-reduction) but the diagnosis is about what the thinking is in service of, not how the thinker feels.
Output Persistence
- Coaching sessions: Save diagnostic output to
{topic}-thinking-audit-{date}.md capturing: identified states, mechanism diagnosis, interventions applied, results observed, and follow-up recommendations.
- Self-monitoring: Results stay in conversation context. No file output unless the user requests a record.
Reasoning Requirements
Extended thinking recommended for:
- Multi-state diagnosis (when multiple GT states may be co-active)
- Cascade analysis across operations (tracing how one operation failure triggers others)
- Complex orientation-capture identification (especially GT5 — Monitor co-option is designed to be invisible from inside)
- Differentiating between similar surface presentations with different mechanisms
Context Management
Base skill: ~3k tokens. Full state definitions: ~5k tokens. Reference documents (references/structural-account.md and references/agent-instructions.md) loaded on-demand when deep theory, anchoring examples, or detailed failure-mode analysis is needed. For routine diagnosis, the SKILL.md alone is sufficient.
Weekly Installs
105
Repository
jwynia/agent-skills
GitHub Stars
38
First Seen
Feb 14, 2026
Security Audits
Gen Agent Trust HubPassSocketPassSnykPass
Installed on
codex98
opencode98
gemini-cli97
github-copilot96
kimi-cli95
amp95
