Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Instruções da origem · Visualização somente leitura
name
linear-meeting-actions
description
Create Linear issues or project updates from Slack context, meeting notes, and files with project and assignee resolution
requires_auth
true
parameters
[{"action":"One of extract, create, approve, or reject. Defaults to create for transcript-to-Linear requests."},{"transcript":"Pasted meeting transcript, notes, or summary text. If omitted, use the current Slack thread or bounded recent channel context and supported attached files."},{"project_hint":"Optional Linear project name or slug mentioned by the user."},{"owner_hint":"Optional Linear assignee name, email, or Slack mention."},{"team_hint":"Optional Linear team name or key mentioned by the user."},{"confirm_create":"Boolean flag used by Slack approval actions for uncertain items."}]
routing
{"use_when":"The user wants Linear issues/tasks or project updates created from a direct request, nearby Slack context, meeting notes, a thread, or attached files.\n","avoid_when":"MLAI points tasks (mlai-points), Linear how-to questions, or tasks without a Linear destination.\n","examples":[{"text":"add a task to linear to fix the login bug","action":"create"},{"text":"send this attached PDF to Linear as tasks","action":"create"},{"text":"add this as a task for me in Linear","action":"create"}],"negative_examples":[{"text":"create a task called fix docs worth 5 points","instead":"mlai-points"},{"text":"how do I write a good linear ticket?","instead":"respond_in_chat"}]}
actions
[{"name":"create","description":"Extract candidate action items (from the message, thread, or files) and create Linear issues.","params":{"project_hint":{"type":"string","description":"Linear project name/slug if mentioned."},"team_hint":{"type":"string","description":"Linear team name/key if mentioned."}}},{"name":"extract","description":"Only extract/preview action items without creating issues yet.","params":{"project_hint":{"type":"string"},"team_hint":{"type":"string"}}},{"name":"approve","description":"Approve previously previewed uncertain items (usually a button follow-up)."},{"name":"reject","description":"Reject previously previewed uncertain items."}]
Linear Meeting Actions Skill
This skill turns pasted Slack meeting transcripts, summaries, supported Slack files, PDFs, DOCX documents, and images into Linear issues. It reads Linear context first, extracts action items, maps each item to the right owner and project, creates high-confidence issues, and asks for Slack review when assignment or project confidence is uncertain.
Capabilities
Parse Slack message/thread text and supported attached files for concrete to-do items.
For top-level commands such as "add this as a task", read a bounded slice of the preceding channel conversation on demand.
Create a Linear issue directly from an explicit Slack command such as "create a to do item in Linear project X and assign to Y".
Create an explicit, high-confidence contextual assignment in one step; keep discussion-derived follow-ups review-only.
Create a Linear project update only when the request affirmatively asks for one and Roo can confidently match the project; honor explicit instructions to skip, omit, or not write an update.
Use the latest Linear project update and recent project issues as context for concise PDF/transcript-derived project updates.
Download and parse .pdf, .docx, .txt, .md, .csv, .png, .jpg, .jpeg, .webp, and non-animated .gif files from the current Slack thread.
Inspect Linear teams, users, active projects, project members, labels, and recent open issues.
Assign new issues to the best matching Linear user.
Attach new issues to the best matching Linear project and team.
Avoid likely duplicate open issues and make retries idempotent from Slack source evidence.
Use the exact gpt-5.6-sol model through the OpenAI Responses API for every Linear inference stage.
Select model reasoning effort from source length, source count, ambiguity, partial work, dependencies, artifacts, and conflicting context; never derive model effort from the XS/S/M/L/XL task label.
Recover a timed-out meeting-note chunk once at smaller scope while keeping the overall extraction bounded and creating no partial Linear issues.
For projects whose current Linear name starts exactly with [Studio], estimate the remaining effort in bounded batches with the dedicated Studio rubric and apply exactly one compatible XS/S/M/L/XL label.
Ask for Slack approval when an action item is useful but not confident enough to create automatically.
Parameters
action: One of extract, create, approve, or reject. Defaults to create.
transcript: Pasted meeting transcript, notes, or summary text. If omitted, inspect the current Slack thread or bounded recent channel context and attached files.
project_hint: Optional Linear project name or slug.
owner_hint: Optional Linear assignee name, email, or Slack mention.
team_hint: Optional Linear team name or key.
confirm_create: Internal boolean used for Slack approval actions.
Workflow
Build the source text from the current Slack thread or, for a top-level contextual reference, at most the configured recent channel window. Resolve speaker profiles, source-local timestamps, and the evidence permalink. Roo does not read other channels or proactively cache channel traffic.
Read Linear before writing:
Teams
Users
Active projects with descriptions/content, linked Slack channel, teams, and true members when the Linear schema supports them
Latest project update for each active project when available
Issue labels
Recent open issues for duplicate detection
For explicit direct issue commands, parse the command itself as the issue source, including project and assignee hints.
Otherwise, extract concrete action items into structured candidates with title, description, owner hint, project terms, due expression/date, explicit-commitment flag, evidence message timestamp, source label, and confidence.
If a project update was affirmatively requested and not explicitly negated, summarize all parsed source chunks, compare against the latest project update context, and create a concise Linear project update for the matched project.
If no concrete action item is found but the user asked to add the thread context to Linear, draft one contextual issue and require Slack approval before creation.
Match owners by explicit requester/self-reference and Slack mention/email first, then unique Linear name/email-prefix match, then display/name similarity.
Match projects by explicit project hint, exact name/slug, linked Slack channel, project description/content/update/recent issues, and true project membership as a tie-breaker.
Resolve relative dates from the evidence message in the configured workspace timezone.
Suppress candidates that are already completed, cancelled, or duplicates.
For each resolved project whose current title starts exactly with [Studio], read its bounded project updates, active work, terminal references, related issues, label registry, and labelled precedents. Estimate candidates in bounded groups of three with structured gpt-5.6-sol Responses API calls.
Estimate remaining work rather than original scope: XS is up to 15 minutes, S is up to 1 hour, M is up to 2 hours, L is up to 3 hours, and XL is up to 5 hours or has substantial uncertainty. Work over 5 hours is XL and should be split.
Start direct cleanup at low, ordinary extraction and source summaries at medium, and contextual inference, project synthesis, and Studio sizing at high. Increase to xhigh when deterministic complexity signals warrant it.
Creation Rules
Do not create an issue without a matched Linear team.
Do not auto-create an issue without a high-confidence assignee and project.
A contextual command can auto-create only when its source contains an explicit assignment/commitment and the assignee, project, team, and due date are unambiguous.
Do not auto-create contextual discussion-thread issues; always request Slack approval first.
Project updates are created immediately only when affirmatively requested, not when the request says not to write one, and the project match is confident.
Apply the compatible meeting-action label if it already exists.
In Studio sizing review or required mode, apply exactly one of Extra Small (XS), Small (S), Medium (M), Large (L), or Extra Large (XL). Fail closed if sizing context, structured output, or the compatible label is unavailable.
Studio sizing shadow mode captures estimates without mutating issue labels or descriptions; review sends every Studio candidate for approval; required may auto-create only a high-confidence, context-sufficient assessment.
Issue descriptions must include the context, evidence snippet, source message permalink/identifiers when available, and a note that Roo generated the issue from Slack context.
Response Style
Keep Slack responses concise. Lead with created Linear links, then list review-needed or skipped items.
Treat each extracted source chunk as one inference source; track total batch chunks separately so a long PDF does not artificially increase every chunk's reasoning effort.
Retry a structured parsing or workflow-contract failure at most once. For a missing Studio structured response, lower reasoning effort and increase the output allowance so the model can emit the full schema; for other validation failures, use the next reasoning level.
If meeting-action inference times out, retry only the failed chunk once at smaller paragraph/page-aligned scope with bounded concurrency. Finish extraction and deduplication before any issue write; if recovery or the total extraction deadline fails, create nothing and say so explicitly.
Discard duplicate or unrecognized optional evidence references without rejecting an otherwise valid Studio assessment. Normalize its rationale to one sentence and, when duration-based, add the rubric's canonical time anchor if the model omitted one.
If a Studio sizing batch still fails its required candidate or rubric fields, retry its candidates individually and preserve successful assessments. Add the exact effort label and normalized one-sentence rationale to the issue description and Slack preview. Never create an eligible Studio issue without exactly one valid effort label.
Send an idempotency fingerprint and preserve a Slack permalink in the Linear issue.
Present uncertain candidates in Slack with Approve and Reject buttons.
Report created project updates, created issues, skipped duplicates, and unresolved items clearly.
