Stackon
build · skills

Turn one good run into a skill the whole team can reuse.

Scribe reads a recorded trace — its spans, model calls, and IO — and distills the generalized procedure into a tight, reusable skill. Attach it to any canvas node and the runner prepends it to that node's system prompt automatically. Claude Skills-compatible, so your hard-won workflows become building blocks.

scribe · distilltrace → skill

recorded trace

plan
review
verdict
cite

4 spans · $0.004

distillsonnet 4.6

skill

Tighten reviewer rubric

  1. 1.Read the diff before the description.
  2. 2.Flag security issues first; cite lines.
  3. 3.Return verdict as JSON { pass, notes }.
# Applied skill → reviewer nodeprepended to system prompt
Trace → reusable skillAttaches to any canvas nodeClaude Skills-compatible

01

Pick a trace, get a skill

Choose a run that went well — a clean code review, a tight research loop, a reviewer rubric that finally caught the right things. Scribe reads the full trace: span names, model choices, tool patterns, and IO previews. It extracts the procedure the run is an instance of and writes it as a concrete, imperative skill — not a vague summary, but numbered steps with the format constraints and verdict shapes preserved verbatim.

scribe · distilltrace → skill

recorded trace

plan
review
verdict
cite

4 spans · $0.004

distillsonnet 4.6

skill

Tighten reviewer rubric

  1. 1.Read the diff before the description.
  2. 2.Flag security issues first; cite lines.
  3. 3.Return verdict as JSON { pass, notes }.
# Applied skill → reviewer nodeprepended to system prompt

02

Applied at the top of the prompt

A skill is a prompt prefix, not a black box. Bind one to a canvas node or roster agent and the runner splices it in as an "# Applied skill" block ahead of the base role, retrieved knowledge, and conversation history — so it's the most prescriptive instruction the agent sees. Every node opts in explicitly; nothing is applied behind your back.

canvas · pr-reviewrunning
Plannerdone
Coderlive
Reviewerqueued
agent.run · 3 spans2 / 3 nodes · streaming

03

Editable, with provenance you can trust

Distillation produces a name, a description, and the instruction body — all editable before you ship them. Each skill keeps its source trace IDs, so you can click straight back to the run it came from. Nothing is fabricated: the skill is grounded in what actually happened in your trace.

trace · run_8c4fok · 742ms · $0.0053
agent.plan742ms
tools.search_code86ms
llm.complete_refactor612ms
tools.edit_file78ms
evals.no_regression54ms
agentllmtooleval5 spans · 3,007 tok

04

Cheap, governed, redaction-aware

Distillation runs on Sonnet 4.6 with adaptive thinking and typically costs about half a cent. It honors your team's Cost budgets, routes through the Compliance redaction layer so PII never reaches the model, and writes a scribe.distill audit event. Skills are governed the same way the rest of Stackon is.

compliance · trust layeraudit-ready
agent.coderAgent run· trace · 8c4f21a12:04:11
u · danaRole changed· member · owner12:04:42
proxyPII redacted· 3 replacements12:05:09
u · renBYOK key rotated· anthropic12:06:30
pii proxy · standardscrubbing

in email dana@acme.io, card 4242 4242 4242 4242

out email <REDACTED:email>, card <REDACTED:credit_card>

anthropic…aF3kopenai…9Qx2
AES-256-GCM

Sonnet 4.6

Distill model

~$0.005 / skill

Typical cost

Claude Skills-compatible

Format

Speed plus trust — prove your agents got better this week.

Scribe is one piece of Stackon, the observability-first workspace for teams running Claude and Codex. Start free and instrument your first run today.