Forsy Trace Skill

Structured traces for agent work experience.

Forsy Trace Skill is an open skill for capturing AI agent workflows as structured, annotated trajectory data.

npx forsy-trace-skill init

It 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.

Forsy 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.

Why structured agent traces matter

Agents increasingly work across tools, files, code, research environments, scientific workflows, legal analysis, product prototyping, and operational tasks.

The final output alone is not enough to understand what happened.

A useful agent work trace should capture:

• what the agent was trying to do
• what context and tools it had
• which actions it took
• what it observed after each action
• where it failed, retried, or corrected course
• what feedback shaped the work
• what artifact or outcome was produced
• what signals could be reused by future agents

Forsy Trace Skill gives those workflows a structured format.

What is included

skill.md
docs/schema.md
schema/forsy_trace_schema_v0_1.json
examples/
dataset/
scripts/

skill.md

The open Forsy Trace Skill.

Use it as an instruction file for agents that need to produce structured traces of completed workflows.

schema/

A JSON Schema for the public trace format.

examples/

A seed set of structured text-based agent work traces across multiple workflow types.

Current examples include:

• molecular docking and computational drug discovery
• agentic product prototyping
• scientific computing
• applied math and code optimization
• legal and policy research
• structured legal drafting
• quantitative Hawkes process estimation
• injection moulding process optimization
• hardware/product planning

Each example is organized as:

examples//
  manifest.json
  trace.json

dataset/

Machine-readable JSONL exports:

manifests.jsonl
traces.jsonl
steps.jsonl
normalization_report.json

scripts/

Utilities for validation and export:

validate_traces.py
build_jsonl_exports.py
normalize_traces.py

Install locally

You can copy the skill and schema into your local agent project:

npx forsy-trace-skill init

By default, this creates:

.forsy/trace-skill/
  skill.md
  schema/
    forsy_trace_schema_v0_1.json

Custom output path:

npx forsy-trace-skill init --out skills/forsy-trace-skill

Overwrite existing files:

npx forsy-trace-skill init --force

The installer only copies local files. It does not call external services, run a harness, or submit traces anywhere.

Trace format

A Forsy trace is a structured record of an agent workflow.

A trace can include:

• trace_id
• schema_version
• trace_mode
• task
• agent_tools
• system_prompt
• agent_config
• steps
• learning
• termination_reason
• final_output
• static_output
• summary
• dataset_summary

Each step can include:

• actor
• action
• tool
• input
• output
• observation
• state change
• feedback
• retry relationship
• causal relationship
• local evaluation signal

See docs/schema.md for the full schema guide.

Using the skill

A typical workflow:

1. Add skill.md to your agent environment.
2. Ask the agent to complete or reconstruct a real workflow.
3. Save the structured trace as trace.json.
4. Validate the trace.
5. Export traces into JSONL for downstream analysis.

Validate examples:

python3 scripts/validate_traces.py

Rebuild JSONL exports:

python3 scripts/build_jsonl_exports.py

What this is useful for

Forsy Trace Skill is designed for:

• agent workflow inspection
• tool-use trajectory analysis
• process-supervision research
• agent evaluation
• failure and retry analysis
• annotated trajectory data construction
• workflow auditability
• reusable agent work experience

The included examples are text-based structured traces that demonstrate the format across different workflow types.

Repository structure

forsy-trace-skill/
  .gitignore
  CITATION.cff
  skill.md
  docs/
    schema.md
  schema/
    forsy_trace_schema_v0_1.json
  raw/
  examples/
  dataset/
    manifests.jsonl
    traces.jsonl
    steps.jsonl
    normalization_report.json
  scripts/
    normalize_traces.py
    validate_traces.py
    build_jsonl_exports.py

Citation

If you use Forsy Trace Skill, please cite the repository using the metadata in CITATION.cff.

License

See LICENSE.