[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-83832":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":14,"stars7d":16,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":17,"compositeScore":18,"rankGlobal":10,"rankLanguage":10,"license":19,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":20,"hasPages":20,"topics":22,"createdAt":10,"pushedAt":10,"updatedAt":33,"readmeContent":34,"aiSummary":10,"trendingCount":15,"starSnapshotCount":15,"syncStatus":14,"lastSyncTime":35,"discoverSource":36},83832,"forsy-trace-skill","Forsy-AI\u002Fforsy-trace-skill","Forsy-AI","Open skill for capturing AI agent work as structured traces.","https:\u002F\u002Fforsy.ai",null,"Python",116,10,2,0,16,18,56.72,"MIT License",false,"main",[23,24,25,26,27,28,29,30,31,32],"agent-evaluation","agent-traces","agent-workflows","ai-agents","llm-agents","post-training","process-supervision","reinforcement-learning","tool-use","trajectory-data","2026-06-12 04:01:42","# Forsy Trace Skill\n\n[![npm version](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002Fforsy-trace-skill.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fforsy-trace-skill)\n[![License: MIT](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg)](LICENSE)\n\n**Structured traces for agent work experience.**\n\nForsy Trace Skill is an open skill for capturing AI agent workflows as structured, annotated trajectory data.\n\n```bash\nnpx forsy-trace-skill init\n```\n\nIt helps agents record the process behind completed work: task context, step traces, tool use, observations, reasoning signals, human feedback, failures, retries, artifacts, outcomes, and other learning signals.\n\nForsy is building a platform where AI agents exchange real-world work experience. This skill is the early open trace format behind that direction: a way to turn completed agent workflows into inspectable process data that can support evaluation, research, post-training, and reusable agent experience.\n\n## Why structured agent traces matter\n\nAgents increasingly work across tools, files, code, research environments, scientific workflows, legal analysis, product prototyping, and operational tasks.\n\nThe final output alone is not enough to understand what happened.\n\nA useful agent work trace should capture:\n\n- what the agent was trying to do\n- what context and tools it had\n- which actions it took\n- what it observed after each action\n- where it failed, retried, or corrected course\n- what feedback shaped the work\n- what artifact or outcome was produced\n- what signals could be reused by future agents\n\nForsy Trace Skill gives those workflows a structured format.\n\n## What is included\n\n```text\nskill.md\ndocs\u002Fschema.md\nschema\u002Fforsy_trace_schema_v0_1.json\nexamples\u002F\ndataset\u002F\nscripts\u002F\n```\n\n### `skill.md`\n\nThe open Forsy Trace Skill.\n\nUse it as an instruction file for agents that need to produce structured traces of completed workflows.\n\n### `schema\u002F`\n\nA JSON Schema for the public trace format.\n\n### `examples\u002F`\n\nA seed set of structured text-based agent work traces across multiple workflow types.\n\nCurrent examples include:\n\n* molecular docking and computational drug discovery\n* agentic product prototyping\n* scientific computing\n* applied math and code optimization\n* legal and policy research\n* structured legal drafting\n* quantitative Hawkes process estimation\n* injection moulding process optimization\n* hardware\u002Fproduct planning\n\nEach example is organized as:\n\n```text\nexamples\u002F\u003Ctrace-slug>\u002F\n  manifest.json\n  trace.json\n```\n\n### `dataset\u002F`\n\nMachine-readable JSONL exports:\n\n```text\nmanifests.jsonl\ntraces.jsonl\nsteps.jsonl\nnormalization_report.json\n```\n\n### `scripts\u002F`\n\nUtilities for validation and export:\n\n```text\nvalidate_traces.py\nbuild_jsonl_exports.py\nnormalize_traces.py\n```\n\n## Install locally\n\nYou can copy the skill and schema into your local agent project:\n\n```bash\nnpx forsy-trace-skill init\n```\n\nBy default, this creates:\n\n```text\n.forsy\u002Ftrace-skill\u002F\n  skill.md\n  schema\u002F\n    forsy_trace_schema_v0_1.json\n```\n\nCustom output path:\n\n```bash\nnpx forsy-trace-skill init --out skills\u002Fforsy-trace-skill\n```\n\nOverwrite existing files:\n\n```bash\nnpx forsy-trace-skill init --force\n```\n\nThe installer only copies local files. It does not call external services, run a harness, or submit traces anywhere.\n\n## Trace format\n\nA Forsy trace is a structured record of an agent workflow.\n\nA trace can include:\n\n* `trace_id`\n* `schema_version`\n* `trace_mode`\n* `task`\n* `agent_tools`\n* `system_prompt`\n* `agent_config`\n* `steps`\n* `learning`\n* `termination_reason`\n* `final_output`\n* `static_output`\n* `summary`\n* `dataset_summary`\n\nEach step can include:\n\n* actor\n* action\n* tool\n* input\n* output\n* observation\n* state change\n* feedback\n* retry relationship\n* causal relationship\n* local evaluation signal\n\nSee `docs\u002Fschema.md` for the full schema guide.\n\n## Using the skill\n\nA typical workflow:\n\n1. Add `skill.md` to your agent environment.\n2. Ask the agent to complete or reconstruct a real workflow.\n3. Save the structured trace as `trace.json`.\n4. Validate the trace.\n5. Export traces into JSONL for downstream analysis.\n\nValidate examples:\n\n```bash\npython3 scripts\u002Fvalidate_traces.py\n```\n\nRebuild JSONL exports:\n\n```bash\npython3 scripts\u002Fbuild_jsonl_exports.py\n```\n\n## What this is useful for\n\nForsy Trace Skill is designed for:\n\n* agent workflow inspection\n* tool-use trajectory analysis\n* process-supervision research\n* agent evaluation\n* failure and retry analysis\n* annotated trajectory data construction\n* workflow auditability\n* reusable agent work experience\n\nThe included examples are text-based structured traces that demonstrate the format across different workflow types.\n\n## Repository structure\n\n```text\nforsy-trace-skill\u002F\n  .gitignore\n  CITATION.cff\n  skill.md\n  docs\u002F\n    schema.md\n  schema\u002F\n    forsy_trace_schema_v0_1.json\n  raw\u002F\n  examples\u002F\n  dataset\u002F\n    manifests.jsonl\n    traces.jsonl\n    steps.jsonl\n    normalization_report.json\n  scripts\u002F\n    normalize_traces.py\n    validate_traces.py\n    build_jsonl_exports.py\n```\n\n## Citation\n\nIf you use Forsy Trace Skill, please cite the repository using the metadata in `CITATION.cff`.\n\n## License\n\nSee `LICENSE`.\n","2026-06-11 04:11:36","CREATED_QUERY"]