synth-thread-to-sop

type	skill
name	synth-thread-to-sop
description	Read a resolved Slack thread markdown export and synthesize a typed memory entry (decision, exception, or workflow) into Meta/. Trigger /synth-thread-to-sop <slack-thread-markdown-file>. Use when a thread captures a one-time decision, a documented deviation, or a repeatable procedure worth filing. Do NOT use for raw Slack ingestion (use the Slack ingest skill for that) or for PR sources (use /synth-pr-to-sop).
argument-hint	<slack-thread-markdown-file> [--vault-root PATH] [--dry-run]
tool_access	["Bash","Read","Write"]
required_inputs	[{"name":"thread_path","type":"string","required":true,"description":"Path to a Slack thread markdown file (e.g. External Inputs/Slack/<channel>/<date>.md)."},{"name":"vault_root","type":"string","required":false,"description":"Vault root. Defaults to current working directory."}]
output_shape	{"format":"markdown-file","fields":{"output_path":"absolute path to Meta/Decisions/<sha8>.md, Meta/Exceptions/<sha8>.md, or Meta/Workflows/<sha8>.md","classification":"decision, exception, or workflow","source_thread_url":"thread permalink or root ts"}}

/synth-thread-to-sop

Turn a resolved Slack thread into a typed memory entry. The script first classifies the thread (decision, exception, or workflow definition) using deterministic signals, then writes a typed file at the right location with the right schema.

When to run

After a thread reaches a resolution worth filing. Examples:

• A thread where the team picks one option over another for a one-time call (decision, episodic)
• A thread where someone explains "we don't follow rule X for client Y" (exception, procedural)
• A thread where someone walks the team through a multi-step procedure (workflow, procedural)

Two synthesis paths

Path A: heuristic (offline, no LLM)

The script classifies the thread by counting decision-language hits, exception-language hits, and step-language hits. Highest-scoring class wins. Good for clearly structured threads. Run:

python3 skills/synth-thread-to-sop/synth.py <thread-path> --vault-root <vault>

Path B: Claude Code session is the LLM

Default and recommended. The script writes a draft. Then Claude reads the source thread, refines the classification if wrong, fills in fields the heuristic missed, and writes the final file. No external LLM API call is made by the script.

Step 1: Locate the thread markdown

The argument is a single Slack thread markdown file. Expected structure (any of these is fine):

• A thread exported by your ingest pipeline at External Inputs/Slack/<channel>/<date>.md
• A markdown file with a series of **user (timestamp):** message lines
• A markdown file with frontmatter that contains thread_url, root_ts, or permalink

The script extracts the thread URL or root timestamp from frontmatter first, then from the body if absent.

Step 2: Run the script

python3 skills/synth-thread-to-sop/synth.py <thread-path> --vault-root <vault>

Output (one of):

• Meta/Decisions/<sha8>.md if classified as a decision
• Meta/Exceptions/<sha8>.md if classified as an exception
• Meta/Workflows/<sha8>.md if classified as a workflow

The <sha8> is the first 8 chars of a sha1 of the thread root ts (or URL if no ts). Idempotent.

Step 3: Refine in-session (optional)

If the classification looks wrong, override with --classify-as decision|exception|workflow. Then read the file and improve fields with the Edit tool.