From 138df2154f82aa9c8f7a11a1d4045406d325d0c1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=85=89=E4=B9=8B=E5=B8=9D=E7=9A=87=E4=BE=A0?= <83931096+MycatQieZi@users.noreply.github.com> Date: Thu, 25 Jun 2026 11:47:58 +0800 Subject: [PATCH 1/3] feat: add rebel skill - project introspection & goal guardian Add a critical observer skill that audits AI-assisted projects across 4 dimensions (architecture, code quality, process compliance, NFR), tracks crisis levels L0-L4, and proposes structured coup interventions when projects go off-track. Features adaptive sampling and YAML config. --- src/content/skills/rebel.md | 87 +++++++++++++++++++++++++++++++++++++ 1 file changed, 87 insertions(+) create mode 100644 src/content/skills/rebel.md diff --git a/src/content/skills/rebel.md b/src/content/skills/rebel.md new file mode 100644 index 0000000..048cf7a --- /dev/null +++ b/src/content/skills/rebel.md @@ -0,0 +1,87 @@ +--- +name: rebel +title: "Rebel - Project Introspection & Goal Guardian" +description: "A critical observer that audits AI-assisted projects across 4 dimensions (architecture, code quality, process compliance, non-functional requirements), tracks crisis levels (L0-L4), and proposes structured 'coup' interventions when projects go off-track. Features micro/standard/deep audits with adaptive sampling, YAML-driven configuration, and user-approved corrective actions." +source: community +author: zhengyin +githubUrl: https://github.com/MycatQieZi/rebel +category: development +tags: + - project-health + - code-quality + - architecture + - technical-debt + - audit + - goal-tracking + - agent-oversight + - methodology +roles: + - developer + - architect +featured: false +popular: false +isOfficial: false +installCommand: | + git clone https://github.com/MycatQieZi/rebel + cp -r rebel/.agents/skills/rebel ~/.qoder/skills/rebel +date: 2026-06-25 +--- + +## Use Cases + +- Monitor long-term project health during AI-assisted coding sessions +- Detect goal drift, architecture decay, and tech debt accumulation early +- Enforce coding standards and process compliance through automated audits +- Trigger structured corrective proposals ("coups") when a project reaches crisis state +- Maintain a historical record of project health trends and audit decisions +- Balance the coding agent's tunnel vision with an independent critical observer + +## Core Capabilities + +- **4-Dimension Assessment**: Architecture Health (30%), Code Quality (25%), Process Compliance (20%), Non-Functional Requirements (25%) — each scored 0-100 with detailed indicator breakdowns +- **5 Crisis Levels (L0-L4)**: Peace, Watch, Warning, Crisis, Emergency — with automatic escalation and de-escalation rules +- **3 Audit Modes**: Micro Audit (silent 2s check after every task), Standard Audit (full 4-dimension review every 5 tasks), Deep Audit (comprehensive scan every 20 tasks or on crisis trigger) +- **Coup Proposal System**: When crisis reaches L3+, generates structured YAML proposals with diagnosis, evidence, impact analysis, and remediation actions — all requiring user approval before execution +- **Adaptive Sampling**: Audit frequency adjusts based on project state — more frequent when issues are found, less frequent when clean +- **Configuration-Driven**: All rules, thresholds, and behaviors defined in `.rebel/config.yaml` — fully customizable per project +- **Historical Awareness**: Tracks health trends, audit history, and coup outcomes across sessions + +## Example + +``` +User: /rebel + +Agent: ═══════════════════════════════════════════════ + Rebel Audit Report (Standard Audit) + ═══════════════════════════════════════════════ + + Crisis Level: L1 (Watch) + + Health Scores: + Architecture Health: 78/100 ↓ + Code Quality: 85/100 → + Process Compliance: 72/100 ↓ + Non-Functional: 88/100 → + + Composite Health Index: 81.1/100 + + Findings: + [Warning] Goal drift detected in recent 5 commits + [Warning] 3 new TODO markers added since last audit + [Info] Naming convention mostly consistent + + Rebel's Voice: + "The project is drifting. 40% of recent commits + don't align with the stated core goal. Time to + recalibrate before the drift becomes the new normal." + ═══════════════════════════════════════════════ +``` + +## Notes + +- The skill is language-agnostic — works with any programming language or framework +- All data is stored locally in `.rebel/` within the project directory +- The Rebel never modifies code autonomously — all changes require explicit user approval via the coup proposal system +- Companion files include `resources/assessment-rules.md` (detailed scoring rubrics), `resources/coup-playbook.md` (execution handbook), and `resources/rebel-philosophy.md` (design rationale) +- The `.rebel/config.yaml` can be customized to adjust weights, thresholds, sampling frequency, and coup types +- Slash commands: `/rebel` (standard audit), `/rebel --deep` (deep audit), `/rebel --status` (health overview), `/rebel --coup` (manual coup), `/rebel --dimension `, `/rebel --history` From 39e5522531b7c6b91406114361b1d92117d7f991 Mon Sep 17 00:00:00 2001 From: MycatQieZi <83931096+MycatQieZi@users.noreply.github.com> Date: Fri, 26 Jun 2026 10:04:21 +0800 Subject: [PATCH 2/3] feat: update rebel skill to v2.0 modular architecture - Add modular assessment with per-module scoring and cascade impact - Add Rebuild mechanism (/rebel --rebuild) for full project restructuring - Add module-level coup system with consumer/provider cascade analysis - Add Stop Hook for automatic state updates after agent response - Update example output to show modular v2.0 format - Add new slash commands: --module, --rebuild, --history - Update companion resources: module-audit-guide.md, rebuild-playbook.md --- src/content/skills/rebel.md | 70 +++++++++++++++++++++++-------------- 1 file changed, 43 insertions(+), 27 deletions(-) diff --git a/src/content/skills/rebel.md b/src/content/skills/rebel.md index 048cf7a..7069261 100644 --- a/src/content/skills/rebel.md +++ b/src/content/skills/rebel.md @@ -1,7 +1,7 @@ --- name: rebel -title: "Rebel - Project Introspection & Goal Guardian" -description: "A critical observer that audits AI-assisted projects across 4 dimensions (architecture, code quality, process compliance, non-functional requirements), tracks crisis levels (L0-L4), and proposes structured 'coup' interventions when projects go off-track. Features micro/standard/deep audits with adaptive sampling, YAML-driven configuration, and user-approved corrective actions." +title: "Rebel - Project Introspection & Goal Guardian (v2.0 Modular)" +description: "A critical observer that audits AI-assisted projects across 4 dimensions with per-module scoring. v2.0 adds modular architecture (auto module discovery, independent module scoring, module-level coups with cascade impact analysis), the Rebuild mechanism (/rebel --rebuild for full project restructuring), and Stop Hook for automatic state updates. Tracks crisis levels (L0-L4), proposes structured coup interventions, and maintains historical health trends." source: community author: zhengyin githubUrl: https://github.com/MycatQieZi/rebel @@ -15,6 +15,8 @@ tags: - goal-tracking - agent-oversight - methodology + - modular-audit + - project-rebuild roles: - developer - architect @@ -32,19 +34,25 @@ date: 2026-06-25 - Monitor long-term project health during AI-assisted coding sessions - Detect goal drift, architecture decay, and tech debt accumulation early - Enforce coding standards and process compliance through automated audits -- Trigger structured corrective proposals ("coups") when a project reaches crisis state +- Trigger structured corrective proposals ("coups") when a project or module reaches crisis state +- Execute full project restructuring via the Rebuild mechanism when strategic direction changes +- Perform per-module health assessment with independent scoring and cascade impact analysis - Maintain a historical record of project health trends and audit decisions - Balance the coding agent's tunnel vision with an independent critical observer ## Core Capabilities +- **Modular Assessment (v2.0)**: Auto-discovers project modules from README, scores each module independently with base dimensions (70%) + extra dimensions (30%), and tracks inter-module dependencies with cascade impact analysis - **4-Dimension Assessment**: Architecture Health (30%), Code Quality (25%), Process Compliance (20%), Non-Functional Requirements (25%) — each scored 0-100 with detailed indicator breakdowns -- **5 Crisis Levels (L0-L4)**: Peace, Watch, Warning, Crisis, Emergency — with automatic escalation and de-escalation rules -- **3 Audit Modes**: Micro Audit (silent 2s check after every task), Standard Audit (full 4-dimension review every 5 tasks), Deep Audit (comprehensive scan every 20 tasks or on crisis trigger) -- **Coup Proposal System**: When crisis reaches L3+, generates structured YAML proposals with diagnosis, evidence, impact analysis, and remediation actions — all requiring user approval before execution +- **5 Crisis Levels (L0-L4)**: Peace, Watch, Warning, Crisis, Emergency — with automatic escalation/de-escalation at both module and project level. ≥2 modules at L3+ auto-escalates to global crisis +- **3 Audit Modes**: Micro Audit (silent 2s check after every task), Standard Audit (full modular review every 5 tasks), Deep Audit (comprehensive scan + dependency graph rebuild every 20 tasks or on crisis trigger) +- **Module-Level Coup System**: L3+ modules trigger targeted coup proposals scoped to the affected module, with cascade impact analysis on consumer/provider modules — all requiring user approval +- **Global Coup System**: ≥2 modules at L3+ or manual `/rebel --coup` triggers a cross-module restructuring proposal +- **Rebuild Mechanism (`/rebel --rebuild`)**: Two-phase full project restructuring — Phase 1: Strategic Proposal (goal re-establishment, tech stack audit, feature accumulation audit, roadmap) → Phase 2: Tactical Implementation Plan (file-level changes, API migration, data migration, dependency changes, test strategy) +- **Stop Hook**: Automatically updates Rebel state after each agent response, ensuring continuous monitoring without manual intervention - **Adaptive Sampling**: Audit frequency adjusts based on project state — more frequent when issues are found, less frequent when clean -- **Configuration-Driven**: All rules, thresholds, and behaviors defined in `.rebel/config.yaml` — fully customizable per project -- **Historical Awareness**: Tracks health trends, audit history, and coup outcomes across sessions +- **Configuration-Driven**: All rules, thresholds, module definitions, and behaviors defined in `.rebel/config.yaml` and `.rebel/modules.yaml` — fully customizable per project +- **Historical Awareness**: Tracks health trends, audit history, module coup outcomes, and global coup outcomes across sessions ## Example @@ -52,28 +60,35 @@ date: 2026-06-25 User: /rebel Agent: ═══════════════════════════════════════════════ - Rebel Audit Report (Standard Audit) + Rebel Audit Report (Modular v2.0) ═══════════════════════════════════════════════ - + Crisis Level: L1 (Watch) + Modules: 4 total | 3 healthy | 1 watch | 0 crisis + + Project Overview: + Composite Health Index: 79.5/100 + Bottleneck Module: auth-service (68/100) + + Module Scores: + + 🟢 user-service (88/100) weight 30% + ├── Arch: 90 Code: 85 Process: 88 NonFunc: 89 - Health Scores: - Architecture Health: 78/100 ↓ - Code Quality: 85/100 → - Process Compliance: 72/100 ↓ - Non-Functional: 88/100 → - - Composite Health Index: 81.1/100 + 🟡 auth-service (68/100) weight 25% + ├── Arch: 62 Code: 71 Process: 65 NonFunc: 74 + └── [Warning] JWT token refresh logic has race condition - Findings: - [Warning] Goal drift detected in recent 5 commits - [Warning] 3 new TODO markers added since last audit - [Info] Naming convention mostly consistent + 🟢 data-layer (82/100) weight 30% + ├── Arch: 80 Code: 84 Process: 79 NonFunc: 85 + 🟢 api-gateway (85/100) weight 15% + ├── Arch: 88 Code: 82 Process: 80 NonFunc: 90 + Rebel's Voice: - "The project is drifting. 40% of recent commits - don't align with the stated core goal. Time to - recalibrate before the drift becomes the new normal." + "auth-service is decaying. The JWT refresh race + condition has been there for 3 sprints. Fix it + before it becomes a security incident." ═══════════════════════════════════════════════ ``` @@ -82,6 +97,7 @@ Agent: ════════════════════════ - The skill is language-agnostic — works with any programming language or framework - All data is stored locally in `.rebel/` within the project directory - The Rebel never modifies code autonomously — all changes require explicit user approval via the coup proposal system -- Companion files include `resources/assessment-rules.md` (detailed scoring rubrics), `resources/coup-playbook.md` (execution handbook), and `resources/rebel-philosophy.md` (design rationale) -- The `.rebel/config.yaml` can be customized to adjust weights, thresholds, sampling frequency, and coup types -- Slash commands: `/rebel` (standard audit), `/rebel --deep` (deep audit), `/rebel --status` (health overview), `/rebel --coup` (manual coup), `/rebel --dimension `, `/rebel --history` +- Module discovery prioritizes README.md; if absent, auto-generates one from project structure +- Companion resource files: `resources/assessment-rules.md` (scoring rubrics), `resources/coup-playbook.md` (execution handbook), `resources/rebel-philosophy.md` (design rationale), `resources/module-audit-guide.md` (module-level audit operations), `resources/rebuild-playbook.md` (rebuild execution guide) +- The `.rebel/config.yaml` can be customized to adjust weights, thresholds, module definitions, sampling frequency, and coup types +- Slash commands: `/rebel` (standard audit), `/rebel --deep` (deep audit), `/rebel --status` (health overview), `/rebel --coup` (manual global coup), `/rebel --module ` (single module audit), `/rebel --rebuild` (full project restructuring), `/rebel --dimension `, `/rebel --history` From 4d2a989cfbee6560f16d74fbe809f6b2c856d4f9 Mon Sep 17 00:00:00 2001 From: MycatQieZi <83931096+MycatQieZi@users.noreply.github.com> Date: Mon, 29 Jun 2026 15:37:31 +0800 Subject: [PATCH 3/3] feat: update rebel skill to v2.1 with Crow pessimist sub-personality - Add dual personality: Rebel (critical observer) + Crow (pessimist) - Add Crow worst-case analysis with 4 risk levels (fatal/high/medium/low) - Add cascading failure analysis across module dependency chains - Add Crow's Warning footnotes on module-level audit reports - Add Crow Micro warnings to micro-audit output - Update example to show v2.1 dual personality format - Add crow-playbook.md to companion resources - Add pessimist-analysis and worst-case tags --- src/content/skills/rebel.md | 42 +++++++++++++++++++++++++------------ 1 file changed, 29 insertions(+), 13 deletions(-) diff --git a/src/content/skills/rebel.md b/src/content/skills/rebel.md index 7069261..2a79740 100644 --- a/src/content/skills/rebel.md +++ b/src/content/skills/rebel.md @@ -1,7 +1,7 @@ --- name: rebel -title: "Rebel - Project Introspection & Goal Guardian (v2.0 Modular)" -description: "A critical observer that audits AI-assisted projects across 4 dimensions with per-module scoring. v2.0 adds modular architecture (auto module discovery, independent module scoring, module-level coups with cascade impact analysis), the Rebuild mechanism (/rebel --rebuild for full project restructuring), and Stop Hook for automatic state updates. Tracks crisis levels (L0-L4), proposes structured coup interventions, and maintains historical health trends." +title: "Rebel + Crow - Project Introspection & Goal Guardian (v2.1)" +description: "A dual-personality critical observer: the Rebel audits AI-assisted projects across 4 dimensions with per-module scoring, while the Crow (乌鸦) sub-personality overlays worst-case analysis to counter vibe coding's optimism bias. v2.1 adds pessimist risk assessment (assumption invalidation, boundary coverage, cascade impact, recovery capability), cascading failure analysis, and Crow's Warning footnotes on every audit." source: community author: zhengyin githubUrl: https://github.com/MycatQieZi/rebel @@ -17,6 +17,8 @@ tags: - methodology - modular-audit - project-rebuild + - pessimist-analysis + - worst-case roles: - developer - architect @@ -37,22 +39,27 @@ date: 2026-06-25 - Trigger structured corrective proposals ("coups") when a project or module reaches crisis state - Execute full project restructuring via the Rebuild mechanism when strategic direction changes - Perform per-module health assessment with independent scoring and cascade impact analysis +- Counter vibe coding optimism bias with automatic worst-case scenario analysis on every audit +- Identify hidden assumptions, untested boundaries, and cascading failure risks before they hit production - Maintain a historical record of project health trends and audit decisions - Balance the coding agent's tunnel vision with an independent critical observer ## Core Capabilities +- **Dual Personality (v2.1)**: The Rebel discovers problems and gives direction; the Crow (乌鸦) predicts disasters and exposes fatal flaws. Two complementary perspectives on every audit +- **Crow Pessimist Sub-Personality**: Automatically overlays worst-case analysis after each Rebel assessment. Uses independent risk levels (🔴 fatal / 🟠 high / 🟡 medium / 🔵 low) with 4 indicators: assumption invalidation, boundary coverage, cascade impact, recovery capability +- **Cascading Failure Analysis**: The Crow traces module dependency chains to predict domino-effect failures — one module's crash path through all its consumers - **Modular Assessment (v2.0)**: Auto-discovers project modules from README, scores each module independently with base dimensions (70%) + extra dimensions (30%), and tracks inter-module dependencies with cascade impact analysis - **4-Dimension Assessment**: Architecture Health (30%), Code Quality (25%), Process Compliance (20%), Non-Functional Requirements (25%) — each scored 0-100 with detailed indicator breakdowns - **5 Crisis Levels (L0-L4)**: Peace, Watch, Warning, Crisis, Emergency — with automatic escalation/de-escalation at both module and project level. ≥2 modules at L3+ auto-escalates to global crisis -- **3 Audit Modes**: Micro Audit (silent 2s check after every task), Standard Audit (full modular review every 5 tasks), Deep Audit (comprehensive scan + dependency graph rebuild every 20 tasks or on crisis trigger) +- **3 Audit Modes**: Micro Audit (silent 2s check after every task, with `[🐦Crow Micro]` warnings for optimism assumptions), Standard Audit (full modular review every 5 tasks), Deep Audit (comprehensive scan + dependency graph rebuild every 20 tasks or on crisis trigger) - **Module-Level Coup System**: L3+ modules trigger targeted coup proposals scoped to the affected module, with cascade impact analysis on consumer/provider modules — all requiring user approval - **Global Coup System**: ≥2 modules at L3+ or manual `/rebel --coup` triggers a cross-module restructuring proposal - **Rebuild Mechanism (`/rebel --rebuild`)**: Two-phase full project restructuring — Phase 1: Strategic Proposal (goal re-establishment, tech stack audit, feature accumulation audit, roadmap) → Phase 2: Tactical Implementation Plan (file-level changes, API migration, data migration, dependency changes, test strategy) - **Stop Hook**: Automatically updates Rebel state after each agent response, ensuring continuous monitoring without manual intervention - **Adaptive Sampling**: Audit frequency adjusts based on project state — more frequent when issues are found, less frequent when clean -- **Configuration-Driven**: All rules, thresholds, module definitions, and behaviors defined in `.rebel/config.yaml` and `.rebel/modules.yaml` — fully customizable per project -- **Historical Awareness**: Tracks health trends, audit history, module coup outcomes, and global coup outcomes across sessions +- **Configuration-Driven**: All rules, thresholds, module definitions, and behaviors defined in `.rebel/config.yaml` and `.rebel/modules.yaml` — fully customizable per project. Crow pessimist dimension configurable via `pessimist` module in config +- **Historical Awareness**: Tracks health trends, audit history, module coup outcomes, global coup outcomes, and Crow disaster predictions across sessions ## Example @@ -60,7 +67,7 @@ date: 2026-06-25 User: /rebel Agent: ═══════════════════════════════════════════════ - Rebel Audit Report (Modular v2.0) + Rebel Audit Report (v2.1 Dual Personality) ═══════════════════════════════════════════════ Crisis Level: L1 (Watch) @@ -74,10 +81,17 @@ Agent: ════════════════════════ 🟢 user-service (88/100) weight 30% ├── Arch: 90 Code: 85 Process: 88 NonFunc: 89 + ├── 🐦Crow: [🔵low] "Session store assumes Redis + │ never partitions. When it does, all logins fail + │ silently. Defense: add circuit breaker, cost: 2h." 🟡 auth-service (68/100) weight 25% ├── Arch: 62 Code: 71 Process: 65 NonFunc: 74 - └── [Warning] JWT token refresh logic has race condition + ├── [Warning] JWT token refresh has race condition + ├── 🐦Crow: [🟠high] "Race condition + traffic spike + │ = mass token corruption at peak. Trigger: >500 + │ concurrent refreshes. Impact: 10K users logged + │ out. Defense: distributed lock, cost: 4h." 🟢 data-layer (82/100) weight 30% ├── Arch: 80 Code: 84 Process: 79 NonFunc: 85 @@ -85,10 +99,11 @@ Agent: ════════════════════════ 🟢 api-gateway (85/100) weight 15% ├── Arch: 88 Code: 82 Process: 80 NonFunc: 90 - Rebel's Voice: - "auth-service is decaying. The JWT refresh race - condition has been there for 3 sprints. Fix it - before it becomes a security incident." + ═══ 🐦Crow's Warning ═══ + Global Cascade Path: auth-service failure → + api-gateway 503 → user-service timeout → full outage. + "Your auth module isn't just decaying — it's a bomb + wired to every other module." ═══════════════════════════════════════════════ ``` @@ -97,7 +112,8 @@ Agent: ════════════════════════ - The skill is language-agnostic — works with any programming language or framework - All data is stored locally in `.rebel/` within the project directory - The Rebel never modifies code autonomously — all changes require explicit user approval via the coup proposal system +- The Crow's warnings always include: trigger conditions, quantified impact, actionable defense suggestions, and estimated defense cost — never empty fear-mongering - Module discovery prioritizes README.md; if absent, auto-generates one from project structure -- Companion resource files: `resources/assessment-rules.md` (scoring rubrics), `resources/coup-playbook.md` (execution handbook), `resources/rebel-philosophy.md` (design rationale), `resources/module-audit-guide.md` (module-level audit operations), `resources/rebuild-playbook.md` (rebuild execution guide) -- The `.rebel/config.yaml` can be customized to adjust weights, thresholds, module definitions, sampling frequency, and coup types +- Companion resource files: `resources/assessment-rules.md` (scoring rubrics), `resources/coup-playbook.md` (execution handbook), `resources/rebel-philosophy.md` (design rationale), `resources/module-audit-guide.md` (module-level audit operations), `resources/rebuild-playbook.md` (rebuild execution guide), `resources/crow-playbook.md` (Crow pessimist analysis methodology) +- The `.rebel/config.yaml` can be customized to adjust weights, thresholds, module definitions, sampling frequency, coup types, and Crow pessimist dimension indicators - Slash commands: `/rebel` (standard audit), `/rebel --deep` (deep audit), `/rebel --status` (health overview), `/rebel --coup` (manual global coup), `/rebel --module ` (single module audit), `/rebel --rebuild` (full project restructuring), `/rebel --dimension `, `/rebel --history`