| name | meeting-to-artifacts |
| description | Turn a meeting transcript/recording into durable, human-reviewed knowledge artifacts (summary, decisions, requirements, tasks, open questions, risks, disagreements, design/ADR, context updates) routed to configured storage; also pre-reads. Use for "recap this meeting", "extract decisions/action items from this transcript", "process yesterday's meetings", "generate a pre-read". |
Meeting → Artifacts
Transcript in → draft artifacts → human approves → route to durable storage → next
occurrence starts from a pre-read. Rationale: issue
#2.
Config
meeting-artifacts.config.yaml at target repo root. Schema: config.schema.json;
example + defaults: config.example.yaml. Missing file → use defaults, tell the user;
never invent a destination outside the current repo.
provider — input source (general pattern, not meeting-specific). Free-form name
(only zoom exercised so far); interface mcp > cli > api (api needs a public
apiDocsUrl). See reference/providers.md.storage — per-artifact destination: filesystem | filesystem-git | mcp | cli.
See reference/routing.md.knowledgeBase — path to the knowledge-management skill's vault root (its
vault.path, default knowledge/); resolves vague mentions to concrete repos/people
and is the context.md routing target.series — recurring-meeting registry (pre-reads, cancellation audit).effectivenessLog — where per-meeting rows accumulate.
Process
1. Ingest
File path or pasted text: use directly. Recording reference: pull via the configured
provider interface. Unconfigured, unsupported platform, or unreachable
interface: say so and ask for a paste/path — never build one-off scraping.
Capture: title, date, attendees, series + occurrence. If in a series, load the
previous occurrence's open items.
2. Classify & gather context
Determine meeting type and project/repo. Cross-reference knowledgeBase before
drafting: owners resolve to handles, services to repos.
3. Infer which artifacts apply
Draft only artifacts with real signal — mapping in reference/artifact-inference.md.
Small-talk-only meeting → short summary.md, nothing else.
4. Draft
Templates in templates/ define each artifact's shape: read for structure, write the
filled-in artifact. Never copy a blank template into a target repo
(the-loop#36) — only filled artifacts and the opt-in config file leave this skill dir.
Rules:
- Every artifact satisfies
reference/knowledge-base-compat.md (front-matter keys,
meeting/<type> tags, relative links, vault-pointable tree) — the
knowledge-management skill ingests these as its meeting-artifact origin.
- Tasks and open questions require owner + due/answer-by date; otherwise demote to
the summary.
- Decisions record reversibility:
easily-reversible | costly-to-reverse.
- Design/architecture agreements use ADR format (
templates/design.md).
- Note in
summary.md which prior open items this meeting resolved.
5. Human review gate — never skip
Present drafts; the user approves/edits/rejects each before anything is written to
its destination. Transcripts capture what was said, not what was agreed. Use
AskUserQuestion on ambiguous tone (sarcasm vs. commitment). User edits are final.
6. Route approved artifacts
Per storage config (reference/routing.md). Defaults:
| Artifact | Default destination |
|---|
summary.md, decisions.md, risks.md, disagreements.md, tasks.md, open-questions.md, context.md | raw/meetings/<date>-<slug>/<name>.md |
requirements.md, design.md | raw/meetings/<date>-<slug>/ |
decisions.md (only if storage.decisions opts into style: decision-log) | one decision-<nnn>.md per decision + index row in the configured dir |
tasks.md (if storage.tasks is mcp/cli) | one tracker issue per task, owner as assignee |
context.md (if knowledgeBase set) | appended to the knowledge base |
open-questions.md also registers against the series entry for the next pre-read.
One commit, message naming meeting date + title.
7. Effectiveness log
Append one templates/meeting-log-entry.md row to effectivenessLog. Scorecard
(templates/scorecard.md, metrics in reference/metrics-and-scorecard.md) only on
request or configured cadence.
8. Pre-read (on request or series cadence)
templates/pre-read.md per reference/pre-read.md: previous occurrence's unresolved
questions + incomplete tasks → "the N things to resolve today." Separate output; never
folded into routing.
Knowledge/code boundary
This skill produces knowledge artifacts and stops there (decision-011). Never chain into
design or implementation work, and never write code where knowledge is maintained. To act
on an artifact, the user points an agent at it as a separate task in the code repo.
Cancellation candidates
A series whose last several occurrences yielded only summary/tasks — no decisions, no
resolved open questions — gets flagged to kill or convert to async. Meta-output: mention
in your reply and the log entry, no per-meeting file.
