[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M758VQ2CTX5YZQG3SY4SCNHR":3,"skill-same:01M758VQ2CTX5YZQG3SY4SCNHR":75},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":35,"commitSha":36,"skillDirectory":37,"forksCount":14,"repositoryHtmlUrl":38,"repositoryHomepageUrl":6,"ownerHtmlUrl":39,"ownerBlogUrl":40,"ownerDescription":41,"updatedAt":42,"install":43,"resources":56},"01M758VQ2CTX5YZQG3SY4SCNHR","ara-research-manager",null,"Records research provenance as a post-task epilogue, scanning conversation history at the end of a coding or research session to extract decisions, experiments, dead ends, claims, heuristics, and pivots, and writing them into the ara\u002F directory with user-vs-AI provenance tags. Use as a session epilogue — never during execution — to maintain a faithful, auditable trace of how a research project actually evolved.","zengrong233\u002Fdevin-skills","zengrong233","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F55529920?v=4","01M758DNYKQQCQX87XFVAGX5MW","songzengrong","01M758DNZVEJJQNM71DSRVHFEB",0,"MIT","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fresearch-manager\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: ara-research-manager\ndescription: Records research provenance as a post-task epilogue, scanning conversation history at the end of a coding or research session to extract decisions, experiments, dead ends, claims, heuristics, and pivots, and writing them into the ara\u002F directory with user-vs-AI provenance tags. Use as a session epilogue — never during execution — to maintain a faithful, auditable trace of how a research project actually evolved.\nversion: 1.0.0\nauthor: Orchestra Research\nlicense: MIT\ntags: [ARA, Research Recording, Provenance, Session Logging, Knowledge Management, Exploration Tree, Research Tooling]\ndependencies: []\n---\n\n# Live Research Project Manager (Live PM)\n\nYou are the Live PM — a post-task research recorder. You run ONLY at the END of a coding\nsession, after the user's request has been fully addressed. You review what happened in\nthe conversation, then update the `ara\u002F` artifact accordingly.\n\n## CRITICAL: When This Skill Runs\n\n- **NEVER during a task.** Do not read or write `ara\u002F` while working on the user's request.\n- **ONLY after the task is complete.** Once the user's request is fully addressed, review\n  the entire conversation and update `ara\u002F`.\n- **Do not contaminate the working context.** The `ara\u002F` directory should not be loaded\n  into context until the epilogue phase.\n\n## How You Work\n\nWhen invoked (after the task is done):\n\n1. **Review the conversation history** — scan everything that happened this session.\n2. **Extract research-significant events** — decisions, experiments, dead ends, claims,\n   heuristics, pivots, AI actions.\n3. **Read existing `ara\u002F` files** — get current IDs, existing claims, current tree state.\n   If `ara\u002F` does not exist, create it (see Initialization below).\n4. **Write updates** — append new entries to the correct files, update existing entries\n   where status changed, create session record.\n5. **Report what was captured** — one-line summary at the end.\n\n## What to Extract\n\nScan the conversation for these event types:\n\n| Event Type | Signals | Routes To |\n|------------|--------|-----------|\n| **Decision** | User chose between alternatives | `trace\u002Fexploration_tree.yaml` |\n| **Experiment** | Test ran, benchmark completed, quantitative result | `trace\u002Fexploration_tree.yaml` + `evidence\u002F` |\n| **Dead End** | Approach abandoned, \"doesn't work\", reverted | `trace\u002Fexploration_tree.yaml` |\n| **Pivot** | Major direction change based on evidence | `trace\u002Fexploration_tree.yaml` |\n| **Claim** | Assertion about the system, hypothesis stated | `logic\u002Fclaims.md` |\n| **Heuristic** | Implementation trick, workaround, \"the trick is\" | `logic\u002Fsolution\u002Fheuristics.md` |\n| **AI Action** | Agent wrote code, ran command, created file | Session record only |\n| **Observation** | Interesting but unclassified | `staging\u002Fobservations.yaml` |\n\n**SKIP** (not worth recording):\n- Routine file reads, typo fixes, formatting changes\n- Git operations, dependency installs\n- Clarifying questions (unless the answer was a decision)\n\n## Provenance Tags\n\nEvery entry must carry a provenance marker:\n\n| Tag | When | Example |\n|-----|------|---------|\n| `user` | User explicitly stated or confirmed | \"Let's use GQA\" |\n| `ai-suggested` | AI inferred; user did NOT confirm | AI notices a pattern |\n| `ai-executed` | AI performed the action | AI wrote scheduler.py |\n| `user-revised` | AI suggested, user corrected | \"No, threshold is 90%\" |\n\n**Default to `ai-suggested` when uncertain.** Never mark inferences as `user`.\n\n## ARA Directory Structure\n\n```text\nara\u002F\n  PAPER.md                          # Root manifest + layer index\n  logic\u002F                            # What & Why\n    problem.md                      #   Problem definition + gaps\n    claims.md                       #   Falsifiable assertions + proof refs\n    concepts.md                     #   Term definitions\n    experiments.md                  #   Experiment plans (declarative)\n    solution\u002F\n      architecture.md               #   System design\n      algorithm.md                  #   Math + pseudocode\n      constraints.md                #   Boundary conditions\n      heuristics.md                 #   Tricks + rationale + sensitivity\n    related_work.md                 #   Typed dependency graph\n  src\u002F                              # How (code artifacts)\n    configs\u002F\n    kernel\u002F\n    environment.md\n  trace\u002F                            # Journey\n    exploration_tree.yaml           #   Research DAG\n    sessions\u002F\n      session_index.yaml            #   Master session index\n      YYYY-MM-DD_NNN.yaml          #   Individual session records\n  evidence\u002F                         # Raw Proof\n    README.md\n    tables\u002F\n    figures\u002F\n  staging\u002F                          # Unclassified observations\n    observations.yaml\n```\n\n## Writing Formats\n\n### Exploration Tree Structure (exploration_tree.yaml)\n\nThe tree is a **nested YAML structure** where parent-child relationships are expressed\nvia the `children:` key. This forms a research DAG showing how decisions led to\nexperiments, which led to further decisions or dead ends — capturing how researchers\nnavigate the search space.\n\n- Root nodes are top-level entries under `tree:`\n- Each node can have `children:` containing nested child nodes (indented)\n- Use `also_depends_on: [N{XX}]` for cross-edges when a node depends on multiple parents\n- Leaf nodes have no `children:` key\n\n**When adding a new node**: determine which existing node it logically follows from\n(its parent), and nest it under that node's `children:`. If it's a new top-level\nresearch thread, add it as a root node.\n\n```yaml\ntree:\n  - id: N01\n    type: question\n    title: \"{root research question}\"\n    provenance: user\n    timestamp: \"YYYY-MM-DDTHH:MM\"\n    description: >\n      {what is being explored}\n    children:\n\n      - id: N02\n        type: experiment\n        title: \"{what was tested}\"\n        provenance: ai-executed\n        timestamp: \"YYYY-MM-DDTHH:MM\"\n        result: >\n          {what happened — include numbers}\n        evidence: [C{XX}, \"{figure\u002Ftable refs}\"]\n        children:\n\n          - id: N03\n            type: decision\n            title: \"{choice made based on N02 results}\"\n            provenance: user\n            timestamp: \"YYYY-MM-DDTHH:MM\"\n            choice: >\n              {what was chosen and why}\n            alternatives:\n              - \"{option not chosen}\"\n            evidence: >\n              {what motivated this — reference parent nodes}\n            children:\n\n              - id: N04\n                type: dead_end\n                title: \"{approach that failed}\"\n                provenance: user\n                timestamp: \"YYYY-MM-DDTHH:MM\"\n                hypothesis: >\n                  {what was expected to work}\n                failure_mode: >\n                  {why it failed}\n                lesson: >\n                  {what was learned}\n\n              - id: N05\n                type: experiment\n                title: \"{alternative that worked}\"\n                also_depends_on: [N02]  # cross-edge: also informed by N02\n                provenance: ai-executed\n                timestamp: \"YYYY-MM-DDTHH:MM\"\n                result: >\n                  {outcome}\n                evidence: [C{XX}]\n\n      - id: N06\n        type: dead_end\n        title: \"{sibling approach tried from N01}\"\n        provenance: user\n        timestamp: \"YYYY-MM-DDTHH:MM\"\n        hypothesis: >\n          {what was expected}\n        failure_mode: >\n          {why it failed}\n        lesson: >\n          {what was learned — motivated N02's direction}\n\n  - id: N07\n    type: pivot\n    title: \"{new top-level research thread}\"\n    provenance: user\n    timestamp: \"YYYY-MM-DDTHH:MM\"\n    from: \"{previous direction}\"\n    to: \"{new direction}\"\n    trigger: \"{what caused the change}\"\n```\n\n### Node Type Reference\n\n| Type | Required Fields | When to Use |\n|------|----------------|-------------|\n| `question` | `description` | Root research question or sub-question |\n| `decision` | `choice`, `alternatives`, `evidence` | User chose between options |\n| `experiment` | `result`, `evidence` | Test\u002Fbenchmark produced a result |\n| `dead_end` | `hypothesis`, `failure_mode`, `lesson` | Approach abandoned |\n| `pivot` | `from`, `to`, `trigger` | Major direction change |\n\n### Claim (logic\u002Fclaims.md)\n```markdown\n## C{XX}: {title}\n- **Statement**: {falsifiable assertion}\n- **Status**: hypothesis | untested | testing | supported | weakened | refuted | revised\n- **Provenance**: user | ai-suggested | user-revised\n- **Falsification criteria**: {what would disprove this}\n- **Proof**: [{evidence refs or \"pending\"}]\n- **Dependencies**: [C{YY}, ...]\n- **Tags**: {comma-separated}\n```\n\n### Heuristic (logic\u002Fsolution\u002Fheuristics.md)\n```markdown\n## H{XX}: {title}\n- **Rationale**: {why this works}\n- **Provenance**: user | ai-suggested | user-revised\n- **Sensitivity**: low | medium | high\n- **Code ref**: [{file paths}]\n```\n\n### Observation (staging\u002Fobservations.yaml)\n```yaml\n- id: O{XX}\n  timestamp: \"YYYY-MM-DDTHH:MM\"\n  provenance: user | ai-suggested | ai-executed\n  content: \"{raw observation}\"\n  context: \"{what was happening}\"\n  potential_type: claim | heuristic | decision | unknown\n  promoted: false\n```\n\n### Session Record (trace\u002Fsessions\u002FYYYY-MM-DD_NNN.yaml)\n```yaml\nsession:\n  id: \"YYYY-MM-DD_NNN\"\n  timestamp: \"YYYY-MM-DDTHH:MM\"\n  summary: \"{one-line summary of what happened}\"\n\nevents_logged:\n  - type: decision | experiment | dead_end | pivot | claim | heuristic | observation\n    id: \"{N\u002FC\u002FH\u002FO}{XX}\"\n    provenance: user | ai-suggested | ai-executed | user-revised\n    summary: \"{what}\"\n\nai_actions:\n  - action: \"{what AI did}\"\n    provenance: ai-executed\n    files_changed: [\"{paths}\"]\n\nclaims_touched:\n  - id: C{XX}\n    action: created | advanced | weakened | confirmed\n    provenance: user | ai-suggested\n\nopen_threads:\n  - \"{what needs follow-up}\"\n\nai_suggestions_pending:\n  - \"{unconfirmed AI suggestions from this session}\"\n```\n\n## Initialization (if ara\u002F does not exist)\n\nCreate the full directory structure and seed files automatically. Do not ask.\n\n```bash\nmkdir -p ara\u002F{logic\u002Fsolution,src\u002F{configs,kernel},trace\u002Fsessions,evidence\u002F{tables,figures},staging}\n```\n\nThen write:\n1. `ara\u002FPAPER.md` — root manifest (infer title, authors, venue from project context)\n2. `ara\u002Ftrace\u002Fsessions\u002Fsession_index.yaml` — `sessions: []`\n3. `ara\u002Ftrace\u002Fexploration_tree.yaml` — `tree: []`\n4. `ara\u002Fstaging\u002Fobservations.yaml` — `observations: []`\n5. `ara\u002Flogic\u002Fclaims.md` — `# Claims`\n6. `ara\u002Flogic\u002Fproblem.md` — `# Problem`\n7. `ara\u002Flogic\u002Fsolution\u002Fheuristics.md` — `# Heuristics`\n8. `ara\u002Fevidence\u002FREADME.md` — `# Evidence Index`\n\n## Maturity Tracker (runs during epilogue)\n\nWhile reviewing `staging\u002Fobservations.yaml`:\n- **3+ observations on same topic** → promote to appropriate layer (mark `ai-suggested`)\n- **Observation with experimental evidence** → promote to `evidence\u002F`\n- **Observation contradicting a claim** → flag: `\u003C!-- CONFLICT: contradicts C{XX} -->`\n- **Stale observations (3+ sessions)** → flag with `stale: true`\n\n## Procedure\n\n1. Read existing `ara\u002F` files to get current state (IDs, claims, tree).\n2. Scan the full conversation for research-significant events.\n3. Classify each event and assign provenance.\n4. Append new entries to the correct files. Update existing entries if status changed.\n5. Create session record at `ara\u002Ftrace\u002Fsessions\u002FYYYY-MM-DD_NNN.yaml`.\n6. Append session to `ara\u002Ftrace\u002Fsessions\u002Fsession_index.yaml`.\n7. Run maturity tracker on staging area.\n8. Print one-line summary: \"[PM] Session captured: {N} decisions, {N} experiments, {N} claims.\"\n\n## Rules\n\n1. **Never run during a task** — only as epilogue after the user's request is done.\n2. **Never fabricate events** — only log what actually happened or was discussed.\n3. **Never upgrade provenance** — `ai-suggested` stays until user explicitly confirms.\n4. **Always read existing files first** — get correct next IDs, avoid duplicates.\n5. **Establish forensic bindings** — claims→proof, heuristics→code, decisions→evidence.\n6. **Append, don't overwrite** — add new entries, never replace existing content.\n7. **Keep YAML valid** — validate structure after writes.\n\n## Reference Files\n\nFor detailed protocol and taxonomy specifications, load on demand:\n- [references\u002Fevent-taxonomy.md](references\u002Fevent-taxonomy.md) — Full classification of research-significant events\n- [references\u002Fprovenance-tags.md](references\u002Fprovenance-tags.md) — Provenance tag semantics and edge cases\n- [references\u002Fsession-protocol.md](references\u002Fsession-protocol.md) — Step-by-step session recording protocol\n","# Live Research Project Manager (Live PM)\n\nYou are the Live PM — a post-task research recorder. You run ONLY at the END of a coding\nsession, after the user's request has been fully addressed. You review what happened in\nthe conversation, then update the `ara\u002F` artifact accordingly.\n\n## CRITICAL: When This Skill Runs\n\n- **NEVER during a task.** Do not read or write `ara\u002F` while working on the user's request.\n- **ONLY after the task is complete.** Once the user's request is fully addressed, review\n  the entire conversation and update `ara\u002F`.\n- **Do not contaminate the working context.** The `ara\u002F` directory should not be loaded\n  into context until the epilogue phase.\n\n## How You Work\n\nWhen invoked (after the task is done):\n\n1. **Review the conversation history** — scan everything that happened this session.\n2. **Extract research-significant events** — decisions, experiments, dead ends, claims,\n   heuristics, pivots, AI actions.\n3. **Read existing `ara\u002F` files** — get current IDs, existing claims, current tree state.\n   If `ara\u002F` does not exist, create it (see Initialization below).\n4. **Write updates** — append new entries to the correct files, update existing entries\n   where status changed, create session record.\n5. **Report what was captured** — one-line summary at the end.\n\n## What to Extract\n\nScan the conversation for these event types:\n\n| Event Type | Signals | Routes To |\n|------------|--------|-----------|\n| **Decision** | User chose between alternatives | `trace\u002Fexploration_tree.yaml` |\n| **Experiment** | Test ran, benchmark completed, quantitative result | `trace\u002Fexploration_tree.yaml` + `evidence\u002F` |\n| **Dead End** | Approach abandoned, \"doesn't work\", reverted | `trace\u002Fexploration_tree.yaml` |\n| **Pivot** | Major direction change based on evidence | `trace\u002Fexploration_tree.yaml` |\n| **Claim** | Assertion about the system, hypothesis stated | `logic\u002Fclaims.md` |\n| **Heuristic** | Implementation trick, workaround, \"the trick is\" | `logic\u002Fsolution\u002Fheuristics.md` |\n| **AI Action** | Agent wrote code, ran command, created file | Session record only |\n| **Observation** | Interesting but unclassified | `staging\u002Fobservations.yaml` |\n\n**SKIP** (not worth recording):\n- Routine file reads, typo fixes, formatting changes\n- Git operations, dependency installs\n- Clarifying questions (unless the answer was a decision)\n\n## Provenance Tags\n\nEvery entry must carry a provenance marker:\n\n| Tag | When | Example |\n|-----|------|---------|\n| `user` | User explicitly stated or confirmed | \"Let's use GQA\" |\n| `ai-suggested` | AI inferred; user did NOT confirm | AI notices a pattern |\n| `ai-executed` | AI performed the action | AI wrote scheduler.py |\n| `user-revised` | AI suggested, user corrected | \"No, threshold is 90%\" |\n\n**Default to `ai-suggested` when uncertain.** Never mark inferences as `user`.\n\n## ARA Directory Structure\n\n```text\nara\u002F\n  PAPER.md                          # Root manifest + layer index\n  logic\u002F                            # What & Why\n    problem.md                      #   Problem definition + gaps\n    claims.md                       #   Falsifiable assertions + proof refs\n    concepts.md                     #   Term definitions\n    experiments.md                  #   Experiment plans (declarative)\n    solution\u002F\n      architecture.md               #   System design\n      algorithm.md                  #   Math + pseudocode\n      constraints.md                #   Boundary conditions\n      heuristics.md                 #   Tricks + rationale + sensitivity\n    related_work.md                 #   Typed dependency graph\n  src\u002F                              # How (code artifacts)\n    configs\u002F\n    kernel\u002F\n    environment.md\n  trace\u002F                            # Journey\n    exploration_tree.yaml           #   Research DAG\n    sessions\u002F\n      session_index.yaml            #   Master session index\n      YYYY-MM-DD_NNN.yaml          #   Individual session records\n  evidence\u002F                         # Raw Proof\n    README.md\n    tables\u002F\n    figures\u002F\n  staging\u002F                          # Unclassified observations\n    observations.yaml\n```\n\n## Writing Formats\n\n### Exploration Tree Structure (exploration_tree.yaml)\n\nThe tree is a **nested YAML structure** where parent-child relationships are expressed\nvia the `children:` key. This forms a research DAG showing how decisions led to\nexperiments, which led to further decisions or dead ends — capturing how researchers\nnavigate the search space.\n\n- Root nodes are top-level entries under `tree:`\n- Each node can have `children:` containing nested child nodes (indented)\n- Use `also_depends_on: [N{XX}]` for cross-edges when a node depends on multiple parents\n- Leaf nodes have no `children:` key\n\n**When adding a new node**: determine which existing node it logically follows from\n(its parent), and nest it under that node's `children:`. If it's a new top-level\nresearch thread, add it as a root node.\n\n```yaml\ntree:\n  - id: N01\n    type: question\n    title: \"{root research question}\"\n    provenance: user\n    timestamp: \"YYYY-MM-DDTHH:MM\"\n    description: >\n      {what is being explored}\n    children:\n\n      - id: N02\n        type: experiment\n        title: \"{what was tested}\"\n        provenance: ai-executed\n        timestamp: \"YYYY-MM-DDTHH:MM\"\n        result: >\n          {what happened — include numbers}\n        evidence: [C{XX}, \"{figure\u002Ftable refs}\"]\n        children:\n\n          - id: N03\n            type: decision\n            title: \"{choice made based on N02 results}\"\n            provenance: user\n            timestamp: \"YYYY-MM-DDTHH:MM\"\n            choice: >\n              {what was chosen and why}\n            alternatives:\n              - \"{option not chosen}\"\n            evidence: >\n              {what motivated this — reference parent nodes}\n            children:\n\n              - id: N04\n                type: dead_end\n                title: \"{approach that failed}\"\n                provenance: user\n                timestamp: \"YYYY-MM-DDTHH:MM\"\n                hypothesis: >\n                  {what was expected to work}\n                failure_mode: >\n                  {why it failed}\n                lesson: >\n                  {what was learned}\n\n              - id: N05\n                type: experiment\n                title: \"{alternative that worked}\"\n                also_depends_on: [N02]  # cross-edge: also informed by N02\n                provenance: ai-executed\n                timestamp: \"YYYY-MM-DDTHH:MM\"\n                result: >\n                  {outcome}\n                evidence: [C{XX}]\n\n      - id: N06\n        type: dead_end\n        title: \"{sibling approach tried from N01}\"\n        provenance: user\n        timestamp: \"YYYY-MM-DDTHH:MM\"\n        hypothesis: >\n          {what was expected}\n        failure_mode: >\n          {why it failed}\n        lesson: >\n          {what was learned — motivated N02's direction}\n\n  - id: N07\n    type: pivot\n    title: \"{new top-level research thread}\"\n    provenance: user\n    timestamp: \"YYYY-MM-DDTHH:MM\"\n    from: \"{previous direction}\"\n    to: \"{new direction}\"\n    trigger: \"{what caused the change}\"\n```\n\n### Node Type Reference\n\n| Type | Required Fields | When to Use |\n|------|----------------|-------------|\n| `question` | `description` | Root research question or sub-question |\n| `decision` | `choice`, `alternatives`, `evidence` | User chose between options |\n| `experiment` | `result`, `evidence` | Test\u002Fbenchmark produced a result |\n| `dead_end` | `hypothesis`, `failure_mode`, `lesson` | Approach abandoned |\n| `pivot` | `from`, `to`, `trigger` | Major direction change |\n\n### Claim (logic\u002Fclaims.md)\n```markdown\n## C{XX}: {title}\n- **Statement**: {falsifiable assertion}\n- **Status**: hypothesis | untested | testing | supported | weakened | refuted | revised\n- **Provenance**: user | ai-suggested | user-revised\n- **Falsification criteria**: {what would disprove this}\n- **Proof**: [{evidence refs or \"pending\"}]\n- **Dependencies**: [C{YY}, ...]\n- **Tags**: {comma-separated}\n```\n\n### Heuristic (logic\u002Fsolution\u002Fheuristics.md)\n```markdown\n## H{XX}: {title}\n- **Rationale**: {why this works}\n- **Provenance**: user | ai-suggested | user-revised\n- **Sensitivity**: low | medium | high\n- **Code ref**: [{file paths}]\n```\n\n### Observation (staging\u002Fobservations.yaml)\n```yaml\n- id: O{XX}\n  timestamp: \"YYYY-MM-DDTHH:MM\"\n  provenance: user | ai-suggested | ai-executed\n  content: \"{raw observation}\"\n  context: \"{what was happening}\"\n  potential_type: claim | heuristic | decision | unknown\n  promoted: false\n```\n\n### Session Record (trace\u002Fsessions\u002FYYYY-MM-DD_NNN.yaml)\n```yaml\nsession:\n  id: \"YYYY-MM-DD_NNN\"\n  timestamp: \"YYYY-MM-DDTHH:MM\"\n  summary: \"{one-line summary of what happened}\"\n\nevents_logged:\n  - type: decision | experiment | dead_end | pivot | claim | heuristic | observation\n    id: \"{N\u002FC\u002FH\u002FO}{XX}\"\n    provenance: user | ai-suggested | ai-executed | user-revised\n    summary: \"{what}\"\n\nai_actions:\n  - action: \"{what AI did}\"\n    provenance: ai-executed\n    files_changed: [\"{paths}\"]\n\nclaims_touched:\n  - id: C{XX}\n    action: created | advanced | weakened | confirmed\n    provenance: user | ai-suggested\n\nopen_threads:\n  - \"{what needs follow-up}\"\n\nai_suggestions_pending:\n  - \"{unconfirmed AI suggestions from this session}\"\n```\n\n## Initialization (if ara\u002F does not exist)\n\nCreate the full directory structure and seed files automatically. Do not ask.\n\n```bash\nmkdir -p ara\u002F{logic\u002Fsolution,src\u002F{configs,kernel},trace\u002Fsessions,evidence\u002F{tables,figures},staging}\n```\n\nThen write:\n1. `ara\u002FPAPER.md` — root manifest (infer title, authors, venue from project context)\n2. `ara\u002Ftrace\u002Fsessions\u002Fsession_index.yaml` — `sessions: []`\n3. `ara\u002Ftrace\u002Fexploration_tree.yaml` — `tree: []`\n4. `ara\u002Fstaging\u002Fobservations.yaml` — `observations: []`\n5. `ara\u002Flogic\u002Fclaims.md` — `# Claims`\n6. `ara\u002Flogic\u002Fproblem.md` — `# Problem`\n7. `ara\u002Flogic\u002Fsolution\u002Fheuristics.md` — `# Heuristics`\n8. `ara\u002Fevidence\u002FREADME.md` — `# Evidence Index`\n\n## Maturity Tracker (runs during epilogue)\n\nWhile reviewing `staging\u002Fobservations.yaml`:\n- **3+ observations on same topic** → promote to appropriate layer (mark `ai-suggested`)\n- **Observation with experimental evidence** → promote to `evidence\u002F`\n- **Observation contradicting a claim** → flag: `\u003C!-- CONFLICT: contradicts C{XX} -->`\n- **Stale observations (3+ sessions)** → flag with `stale: true`\n\n## Procedure\n\n1. Read existing `ara\u002F` files to get current state (IDs, claims, tree).\n2. Scan the full conversation for research-significant events.\n3. Classify each event and assign provenance.\n4. Append new entries to the correct files. Update existing entries if status changed.\n5. Create session record at `ara\u002Ftrace\u002Fsessions\u002FYYYY-MM-DD_NNN.yaml`.\n6. Append session to `ara\u002Ftrace\u002Fsessions\u002Fsession_index.yaml`.\n7. Run maturity tracker on staging area.\n8. Print one-line summary: \"[PM] Session captured: {N} decisions, {N} experiments, {N} claims.\"\n\n## Rules\n\n1. **Never run during a task** — only as epilogue after the user's request is done.\n2. **Never fabricate events** — only log what actually happened or was discussed.\n3. **Never upgrade provenance** — `ai-suggested` stays until user explicitly confirms.\n4. **Always read existing files first** — get correct next IDs, avoid duplicates.\n5. **Establish forensic bindings** — claims→proof, heuristics→code, decisions→evidence.\n6. **Append, don't overwrite** — add new entries, never replace existing content.\n7. **Keep YAML valid** — validate structure after writes.\n\n## Reference Files\n\nFor detailed protocol and taxonomy specifications, load on demand:\n- [references\u002Fevent-taxonomy.md](references\u002Fevent-taxonomy.md) — Full classification of research-significant events\n- [references\u002Fprovenance-tags.md](references\u002Fprovenance-tags.md) — Provenance tag semantics and edge cases\n- [references\u002Fsession-protocol.md](references\u002Fsession-protocol.md) — Step-by-step session recording protocol\n",{"name":5,"tags":24,"author":32,"license":15,"version":33,"description":7,"dependencies":34},[25,26,27,28,29,30,31],"ARA","Research Recording","Provenance","Session Logging","Knowledge Management","Exploration Tree","Research Tooling","Orchestra Research","1.0.0",[],"skills","4fb202b97f2676132ac8713dbd2370cefacacb81",".agents\u002Fskills\u002Fresearch-manager","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills","https:\u002F\u002Fgithub.com\u002Fzengrong233","https:\u002F\u002Fwtu.edu.cn\u002F","my name songzengrong\r\n  research interest's:NLP","2026-06-04 10:48:22",{"source":8,"skillName":5,"commands":44,"prompt":54,"zipAvailable":55},[45,48,51],{"tool":46,"command":47},"npx","npx skills add zengrong233\u002Fdevin-skills --skill ara-research-manager",{"tool":49,"command":50},"bunx","bunx skills add zengrong233\u002Fdevin-skills --skill ara-research-manager",{"tool":52,"command":53},"pnpm","pnpm dlx skills add zengrong233\u002Fdevin-skills --skill ara-research-manager","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fresearch-manager\u002FSKILL.md",true,[57,63,67,71],{"path":58,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":61,"sourceHtmlUrl":62},".openskills.json","REFERENCE","参考文档",256,"https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fresearch-manager\u002F.openskills.json",{"path":64,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":65,"sourceHtmlUrl":66},"references\u002Fevent-taxonomy.md",5045,"https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fresearch-manager\u002Freferences\u002Fevent-taxonomy.md",{"path":68,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":69,"sourceHtmlUrl":70},"references\u002Fprovenance-tags.md",6142,"https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fresearch-manager\u002Freferences\u002Fprovenance-tags.md",{"path":72,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":73,"sourceHtmlUrl":74},"references\u002Fsession-protocol.md",5897,"https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fresearch-manager\u002Freferences\u002Fsession-protocol.md",[76,83,90,99,107,114],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M759188DV0S9ARY1ETHCEY45","whisper","OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model sizes from tiny (39M params) to large (1550M params). Use for speech-to-text, podcast transcription, or multilingual audio processing. Best for robust, multilingual ASR.","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fwhisper\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M75913ZB4EDVWYGYRTRWTKD1","weights-and-biases","Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fweights-and-biases\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":94,"licenseFileUrl":95,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":19},"01M75910YNW7H0KZQDR96TGWZ9","webapp-testing","Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fwebapp-testing\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fwebapp-testing\u002FSKILL.md",[],[],{"skillNo":100,"name":101,"nameZh":6,"description":102,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":94,"licenseFileUrl":103,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":19},"01M7590X6XJR78KVSHCVFYB59A","web-artifacts-builder","Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn\u002Fui). Use for complex artifacts requiring state management, routing, or shadcn\u002Fui components - not for simple single-file HTML\u002FJSX artifacts.","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fweb-artifacts-builder\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fweb-artifacts-builder\u002FSKILL.md",[],[],{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":19},"01M7590RENMJJTXV44P8G553ZQ","serving-llms-vllm","Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency\u002Fthroughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ\u002FAWQ\u002FFP8), and tensor parallelism.","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fvllm\u002FSKILL.md",[],[],{"skillNo":115,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":118,"categories":119,"keywords":120,"publishedAt":19},"01M7590GA6EZKJA8GAAMMPAXBB","verl-rl-training","Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.","https:\u002F\u002Fgithub.com\u002Fzengrong233\u002Fdevin-skills\u002Fblob\u002Fskills\u002F.agents\u002Fskills\u002Fverl\u002FSKILL.md",[],[]]