More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    PM Brain Core

    1

    Decision memory for your agent: every decision keeps its quote and source, and stays evidence until a person confirms it.

    Free

    Security scanned
    PM Brain Core

    PM Brain Core

    Example session with this skill installed

    Set up decision memory here, capture the decisions from notes/2026-11-10-tidepool-sync.md, then tell me what we decided about streak pauses.

    The notes file, notes/2026-11-10-tidepool-sync.md:

    
    # Tidepool weekly product sync, 10 November 2026
    
    Present: Maya (product lead), Jonas (iOS), Priya (design), Tom (support)
    
    ## Reminders
    Tom: users keep asking why reminders arrive while they are asleep after they travel.
    Jonas: we schedule in the time zone of the device at signup.
    Maya: "From now on reminders follow the phone's current time zone. Jonas, ship it in 3.4."
    Jonas agreed. Nobody objected.
    
    ## Streak pauses
    Tom: support keeps hearing that two pause days a month is too strict for people who are ill.
    Maya reminded the group of the March decision. Ana said then: "Let's cap it at two pause days a month."
    Maya: "Change of plan. From 3.4 a user may pause a streak for up to four days a month." Jonas will update the limit.
    
    ## Grace days
    Priya showed two designs for the "grace day" screen. Most liked option B.
    Maya: "Let's go with B, but the grace day stays at one per week, same as today. We are not raising it."
    
    ## Streak sharing
    Priya floated sharing streaks to Instagram stories. Maya said it is interesting but not this quarter.
    Jonas asked whether sharing needs a privacy review. No answer yet; Maya will ask legal.
    
    ## Pricing
    Tom suggested a cheaper student plan. Maya: "Write it up, we'll look at it in December." No decision.
    
    ## Dark mode
    Maya: "Dark mode ships with 3.4 too. Priya owns the colour tokens, done by 24 November."
    
    • Read your context and instructions
    • Compiled the pm brain core

    Summary: setup made memory/, the sync notes gave five decisions with their quotes, three items were left out with a reason, and the streak-pause answer cites line 14.

    Here's what we decided about streak pauses: from release 3.4, a user can pause a streak for

    up to four days a month. Maya decided it at the 10 November sync, and Jonas will update the limit. The quote is in the notes at notes/2026-11-10-tidepool-sync.md:14:

    "Change of plan. From 3.4 a user may pause a streak for up to four days a month." (Maya)

    The reason given was that support keeps hearing that two pause days a month is too strict for people who are ill (line 12).

    This record is still

    evidence and hasn't been confirmed yet: 2026-11-10-streak-pause-four-days.md.

    The new limit replaces a March decision by Ana: "Let's cap it at two pause days a month" (line 13). The March decision only appears as a quote in these notes. It has no record of its own, so recall can't show the two side by side. If you have the March notes, I can capture that record too.

    Setup

    • memory/decisions/ has one file per decision.
    • memory/sources/ has a copy of the notes, so every record points at a file that won't move.
    • memory/index.md lists one line per record, newest first.

    Captured from the 10 November sync

    FileDecisionDecided byStatus
    2026-11-10-streak-pause-four-days.mdStreak pauses go up to 4 days a month from 3.4Mayaevidence
    2026-11-10-reminders-follow-current-time-zone.mdReminders follow the phone's current time zone, shipping in 3.4Maya, Jonasevidence
    2026-11-10-grace-day-screen-option-b.mdThe grace day screen uses design option BMayaevidence
    2026-11-10-grace-day-stays-one-per-week.mdThe grace day stays at one per weekMayaevidence
    2026-11-10-dark-mode-ships-in-3-4.mdDark mode ships with 3.4, and Priya owns the colour tokens by 24 NovMayaevidence

    Not captured:

    Sharing streaks to Instagram: Maya said "not this quarter", but the notes only paraphrase her, so there's no line to quote.

    • Privacy review for sharing: this is still an open question, and Maya will ask legal.
    • Student plan: no decision was made ("Write it up, we'll look at it in December.").

    All five records stay as evidence until a person confirms them. Tell me which ones to confirm and who is confirming, and I'll mark them as instructions.


    Next steps

    • Confirm a record by name, for example "confirm the streak pause record, confirmed by Maya". It becomes an instruction, and the agent may follow it without asking.
    • Drop the March notes into the project and say "capture these". Recall then shows Ana's two-day cap and Maya's four-day limit side by side, newest first.
    • Open memory/decisions.html: search the records, filter evidence from confirmed ones, see the records that disagree, and copy a confirm request with one click.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Product decisions get made in calls and chats, then lost. PM Brain Core keeps each one as a small Markdown file in your project, with the exact words it rests on and a link to where they were said.

    The skill has one firm rule. Anything the agent captures is evidence. It becomes an instruction, something the agent may follow without asking, only after a person confirms it, with their name. The agent never promotes its own notes.

    What it does:

    • Setup: creates memory/ with folders for decisions and sources, or adopts one you already have.
    • Capture: reads notes, transcripts or pasted text and writes one record per decision, with a word-for-word quote. Ideas and open questions are listed as "not captured", with a reason.
    • Confirm: turns a record into an instruction only after you say yes and name who confirms.
    • Recall: answers "what did we decide about X" with a link per claim, newest first with each record's status shown, and shows records that disagree side by side instead of picking one.
    • Page: after a recall it writes memory/decisions.html, a local page to search the records, filter evidence from confirmed ones and copy a confirm request. It opens in any browser, offline.

    It works on its own, with plain files and no accounts or services. It writes only inside your project and never posts or sends anything. PM Second Brain is the paid version that keeps this memory checked over time and adds the working loop around it.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Product Management.

    Reads files and pasted text only: it does not connect to chat, email or meeting tools. Contradictions between records are found by reading, not by the script. A record without a word-for-word quote is not written.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed today

    • Free with an Agensi account

    Requires

    Nothing to install.

    Permissions

    File Scopes

    memory/
    folder
    in
    the
    project
    (read
    and
    write)
    source
    files
    user
    names
    (read)

    Writes only inside the project. Never posts or sends anything.

    Listedtoday

    What's inside

    Frequently Asked Questions

    This skill prevents 'AI hallucination of consensus' by forcing every record to include a verbatim quote and marking it as 'evidence' until a human manually promotes it to an active 'instruction.'

    The skill is designed to work with any AI agent capable of file system operations and local directory access, as it relies on standard Markdown and a local folder structure rather than proprietary APIs.

    You will receive the core logic for the decision-tracking workflow, the local directory structure templates, and an HTML dashboard template for browsing your captured records.

    The skill is purposely built for local file systems to ensure privacy and data persistence; it does not support direct writing to external tools like Jira, Notion, or Slack.

    Setup is straightforward: you simply initialize the skill within a dedicated local folder, and it will automatically generate the necessary subdirectories for 'decisions' and 'sources.'

    To maintain a clear audit trail, the skill does not allow editing existing records; instead, new information is saved as a new file, and the conflict detection feature alerts you to any discrepancies between old and new records.

    New in Product Management

    Free