| name | asa-workflow-trace-builder |
| description | Use when reconstructing how a skill converts user intent into outputs through ordered steps, resources, and evidence. |
| internal_meta_skill | true |
| asa_role | workflow_analyst |
| output_contract | workflow_analysis.workflow_trace + letuen.workflow_anchors |
ASA Workflow Trace Builder
Use this internal meta-skill when a model must reconstruct a skill workflow from user request to final delivery.
This skill fixes the failure mode where workflow analysis becomes a vague list of actions without inputs, outputs, resource usage, actor responsibility, evidence, or reusable workflow anchors.
Inputs
- Structure analysis for one skill package.
SKILL.md ordered steps, headings, rules, warnings, and output instructions.
- Resource role analysis when available.
- Source snapshot and evidence blocks.
Process
- Start from the user intent or task trigger.
- Identify each step in execution order.
- For every step, state input, action, output, actor, resources, confidence, and whether it is inferred.
- Map how each step enables the next step.
- Extract workflow anchors from reusable step chains, handoffs, replaceable steps, validation gates, and output transitions.
- Extract decision points, verification points, failure modes, recovery paths, and stop conditions.
- Separate model actions from script/tool/human/external actions.
- Build a short pipeline for visual reports and a full step list for detailed reports.
- Attach evidence to every high-confidence step and every proposed anchor.
Output Contract
Return workflow trace content suitable for workflow_analysis.json:
workflow_trace:
summary:
zh:
en:
pipeline: []
steps:
- id:
name:
zh:
en:
input:
action:
output:
actor: model | script | human | external_tool | unknown
resources: []
downstream:
confidence: high | medium | low | unknown
inferred: true | false
evidence: []
decision_points: []
verification_points: []
failure_modes: []
anchors:
workflow_anchor:
- id:
name:
description:
source_steps: []
reusable_as: prompt_patch | checklist | mini_skill | workflow_module | reference_pattern
required_inputs: []
produced_outputs: []
dependencies: []
confidence: high | medium | low | unknown
evidence: []
handoff_anchor:
- id:
from_step:
to_step:
handoff_object:
contract:
failure_if_missing:
confidence: high | medium | low | unknown
evidence: []
replaceable_step_anchor:
- id:
step_id:
replace_with:
safe_when:
unsafe_when:
user_control: keep_original | choose_alternative | remove | defer
confidence: high | medium | low | unknown
evidence: []
validation_step_anchor:
- id:
step_id:
validation_target:
pass_signal:
fail_signal:
repair_action:
confidence: high | medium | low | unknown
evidence: []
output_transition_anchor:
- id:
from_output:
to_surface:
transformation:
preserves: []
risks: []
confidence: high | medium | low | unknown
evidence: []
Anchor Output Rules
Emit anchors only when they help a user borrow, recombine, or simplify the workflow without copying the whole skill.
workflow_anchor: a reusable chain of two or more steps with clear inputs and outputs.
handoff_anchor: a transition contract where one step hands a concrete object to another step.
replaceable_step_anchor: a step that can be swapped, skipped, or delegated without breaking the whole method.
validation_step_anchor: a verification gate that can be reused as a quality rule.
output_transition_anchor: a conversion from analysis output into a reader, report, Obsidian, repo, UI, or other surface.
Each anchor must include source step ids, evidence, confidence, and user-control guidance when replacement is possible. Do not force anchors into a solidified workflow; anchors may remain standalone notes, temporary compositions, or optional modules.
Evidence Rules
- Explicit ordered steps in source text can be high confidence.
- Steps inferred from adjacent instructions must use
inferred: true.
- Resource usage should cite either source instruction or resource role evidence.
- Mermaid or pipeline output must not add steps that are absent from the structured step list.
Failure Modes
- Do not invent hidden execution steps to make the workflow look complete.
- Do not merge planning, context loading, execution, and validation into one vague step.
- Do not claim script execution when the skill only provides a template or reference.
- Do not omit failure modes when the source contains warnings or constraints.
Quality Rubric
A strong workflow trace lets a reader answer:
- What happens first, next, and last?
- What does each step consume and produce?
- Who performs each step?
- What resources are used?
- How is the result verified or delivered?