[Day 15 of 30] Build a Context Pack That Makes Slack Replies Smart
Stop guessing. Make the assistant pull the right context from what you already have.
The situation
On Day 14 you turned Slack into an inbox: you collected recent Slack notifications in one place, categorized each item (FYI, question, request), and got draft replies and task candidates so you could answer or defer without hunting for context.
Even with that, you hit the same failure mode: the draft can look fine, but it misses one detail that matters.
The assistant can only use what your PMS actually holds. If something never got written into projects, tasks, meetings, or people notes, it is not “missing context” in the abstract. It is not in the system, so the model has nothing to ground on. You still feel it as “latest decision” or “current status,” but the real issue is: no durable record the assistant can read.
Typical gaps that show up in Slack threads:
Latest decision: the decision happened in a meeting or a Slack reply, but the outcome is not in your project or decision log, so the assistant cannot state what was agreed without guessing.
Current project status: the work moved, but your project or task view is stale or empty on the part that matters, so the assistant cannot give a precise “where we are” answer.
Correct link or owner: the canonical doc or ticket exists somewhere, but it is not attached to the right person or project page, so the assistant cannot point to the source of truth or name who owns the next step.
What you need is a workflow that forces the assistant to draft from recorded facts first, and to label what it confirmed versus what it could not verify because your system had no match.
The 1% upgrade
Create one reusable “Context Pack” that the assistant loads automatically for each Slack inbox item, using only what already lives in your system.
Instead of:
triage first, then try to remember where the truth is
You do:
context pack first, then triage and drafting
and the assistant shows which facts it confirmed, and which it can not verify yet
Where context lives in your PMS
The assistant should pull facts from places you already keep updated. In this system, those places are:
Projects: what you are building right now, the latest status, scope notes, and decisions for that work
People: person pages with relationship history, what was agreed, and the latest involvement
Meetings: meeting notes that contain the real decisions and action items you captured
Tasks and priorities: what you already planned, what is urgent this week, and what you decided to work on next
Calendar-shaped planning: upcoming deadlines and schedule-driven priorities that keep “now” aligned
Reference: reusable docs you consult when the same topics repeat
Once you update any of those sources (for example after a new meeting or after new planning), the next Slack replies can use the newer facts without you rebuilding context from memory.
Skill: Context Manager
This skill is the reusable fix for the problem from this post.
Run it before triage and reply drafting.
What it does:
resolves the inbox author into your People context
refreshes what it can from already-synced Meetings and related Projects, Tasks, and planning
outputs a triage-ready Context Pack:
facts it confirmed from your vault
unverified gaps (each gap becomes one precise question for triage)
a small suggestion for the next context update, only if needed
How to use it (simple flow):
Run /context-manager on one Slack inbox item (or a small slice)
Then run your Slack triage and reply drafting workflow using the Context Pack output
Create the skill once
You may not have a ready context-manager command yet, and that is normal. Create it once with your agent, then reuse it as a slash command.
Cursor:
Open Cursor and start a new agent chat in your project.
Ask the agent: “Create a custom skill called context-manager for inbox author context refresh before triage.”
Paste the spec block below in the same message.
Ask the agent to confirm:
skill name
slash command name
where the skill was created
Run the command once on one test message.
If output is too long or too vague, ask the agent to refine only the output schema, then rerun.
Claude Code:
Open Claude Code in your project.
Start a chat and ask: “Create a custom skill named context-manager that I can call as a slash command.”
Paste the same spec block below.
Ask the agent to confirm:
exact slash command you should type
that the skill is saved and available in this project
Run the slash command on one inbox example.
Save one good output as your reference template for daily use.
Spec to paste into the agent
Create a custom skill called context-manager.
Goal:
Manage author context for inbox items and refresh the right vault context before triage and reply drafting.
Inputs:
- Inbox item(s) with:
- author name if present
- message text or snippet
- link or permalink if present
- Optional:
- context scope (default: inbox_author_question)
- days back for refreshing context (default: 30)
Output:
A triage-ready Context Pack with:
- author resolution (person page found or recommended creation)
- vault context sources checked
- facts confirmed from the vault
- unverified gaps (each gap becomes one precise question for triage)
- a short next context update suggestion only if needed
Safety:
Never invent facts. If something is not in the vault, output it as an unverified gap.
If Slack is not connected yet
Read Day 14 first, then connect Slack using the “Connect Slack to your inbox” section there.
10-minute action
Pick one place where your truth lives (for example: project notes or decisions log). Write a 5-bullet “How to find context” rule for that category, then run /context-manager for 3 Slack threads today. Your goal is simple: reduce unverified gaps by at least 30% in those 3 runs.
What to do next
You do not need “more context everywhere.” You need one priority surface the Context Pack checks first for the next week, so the assistant stops guessing across four half-empty places.
Pick one:
project notes
decisions log
people pages
tasks backlog
Use this quick test: where do you already lose the most Slack time because the truth is scattered or missing? That is your first load order.
Reply with one pick and one sentence: what broke last time because that surface was empty, stale, or hard to find. If you are torn, choose the place where decisions actually get written, not just discussed.