Blameless Postmortem Writer

    1

    Incident postmortem writer: turns your logs, chat, alerts and deploy history into a blameless postmortem with a sourced timeline, impact, contributing factors and action items.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Blameless Postmortem Writer

    Example session with this skill installed

    Write a blameless postmortem for the checkout outage on 14 March. Evidence: the deploy log, the alert history and the engineering chat excerpt.

    • Read your context and instructions
    • Compiled the blameless postmortem writer

    Illustrative. The incident and the evidence are invented; this shows the structure and depth the skill produces.

    Summary

    • A configuration change raised a connection pool size above the database limit, and checkout attempts failed for 31 minutes.
    • About 18% of checkout attempts returned errors [INFERRED from the error-rate graph; exact count UNKNOWN].
    • Detection came from a customer report, 12 minutes after the first errors; no alert fired.
    • Rolling back the configuration fixed it.
    • Most important action: alert on database connection saturation and validate pool settings against the database limit before deploy.

    Impact

    ItemValueTag
    Duration14:02 to 14:33 UTC (31 min)CONFIRMED (deploy log, alert history)
    Share of checkout attempts failingabout 18% at peakINFERRED (graph description)
    Data lossNone foundUNKNOWN (no payment reconciliation in the evidence)
    Time to detect12 minCONFIRMED (first error 14:02, first report 14:14)
    Time to mitigate17 min after acknowledgementCONFIRMED

    Timeline (UTC)

    TimeEventSourceTag
    13:58Configuration change deployed: pool size 40 to 120 per instancedeploy logCONFIRMED
    14:02First 5xx responses on the checkout APIalert historyCONFIRMED
    14:02 to 14:14No alert fires; error rate climbsalert historyCONFIRMED
    14:14Support relays a customer report of failed paymentschatCONFIRMED
    14:16On-call engineer acknowledges and starts on the application logschatCONFIRMED
    14:21Database shows "too many connections"; the team suspects trafficchatCONFIRMED
    14:24Traffic comparison shows no unusual volumechatCONFIRMED
    14:27The errors are linked to the 13:58 deploychatCONFIRMED
    14:31Configuration rolled backdeploy logCONFIRMED
    14:33Error rate back to normalgraph descriptionINFERRED

    Gap: 14:02 to 14:14 has no human action in the evidence. Whether anyone saw the dashboard is UNKNOWN.

    Contributing factors

    Trigger: pool size per instance (120) times the number of instances exceeded the database connection limit. [INFERRED: instance count and limit are only described as "several" and "a few hundred"]

    • Latent condition: nothing in the deploy process compares pool settings with the database limit.

    Detection gap: no alert watches connection saturation or the checkout error rate; detection depended on a customer report.

    Response gap: the first hypothesis (traffic) was reasonable given the error message and took three minutes to rule out; the deploy history was not on the first dashboard opened.

    What went well / badly / lucky

    • Well: acknowledgement within two minutes of the report; the rollback was one step.
    • Badly: 12 minutes undetected; change history not visible next to the error graph.
    • Lucky: it happened in a low-traffic hour.

    Action items

    #ActionTypeOwner rolePriorityVerification
    1Alert when database connections exceed 80% of the limit for 2 minutesdetectplatform on-call leadP0Alert fires in a staging load test
    2Pre-deploy check comparing pool size times instance count with the database limitpreventplatform engineerP1The check blocks a test change over the limit
    3Show recent deploys on the checkout error dashboardmitigateobservability ownerP1Dashboard shows the last 24 hours of deploys
    4Checkout error-rate alert at 2% for 3 minutesdetectpayments team leadP0Alert fires in a staging fault test

    Open questions

    • Exact instance count and database connection limit at the time.
    • Whether any payments were captured but not recorded (run the reconciliation report for 14:00 to 14:35).
    • Whether anyone saw the error rate between 14:02 and 14:14.

    Next steps

    • Confirm the open questions with the owners named above.
    • Share the draft with the people involved before it is published.

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

    About this skill

    Write the postmortem your team will actually use. Give it the evidence from an outage or failed release (log excerpts, chat messages, alert history, deploy history, a ticket) and it produces a blameless review: a timeline in one time zone where every row cites its source, the impact with its numbers and how they were obtained, contributing factors instead of a single root cause, what went well, what went badly and where you were lucky, and action items with a type, an owner role, a priority and a way to verify each one.

    What makes it different from a generic prompt: every statement is tagged CONFIRMED, INFERRED or UNKNOWN, so gaps are shown instead of filled with guesses. The wording is checked for blame (names, "failed to", "human error"). It computes time to detect, acknowledge, mitigate and resolve when the evidence allows, and it turns vague actions like "improve monitoring" into measurable ones. It also covers thin evidence, conflicting timestamps and a separate customer-facing summary.

    What it reads and does: it works only from the evidence you paste or point it to. It runs no scripts and no commands, makes no network connection and contacts no one.

    Honest limits: it cannot verify your evidence, and the result is only as good as what you give it. It is not for live incident response, for deciding fault, or for legal or compliance statements.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for DevOps & Cloud.

    Works only from the evidence you give it and cannot verify that evidence. Not for live incident response, for deciding fault, or for legal or compliance statements. Quality depends on the quality of the logs, chat and timestamps provided.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 13 days ago

    Requires

    Nothing to install.

    Permissions

    Read Files

    File Scopes

    <the
    evidence
    files
    you
    name>

    A prompt-only skill (SKILL.md). It reads only the evidence you paste or name; it runs no commands, makes no network connection and reads no environment variables.

    Listed13 days ago

    What's inside

    Frequently Asked Questions

    Anything that shows what happened and when: log excerpts, chat messages, alert history, deploy or change history, a ticket. It asks at most three questions about gaps that would change the conclusions, and marks the rest UNKNOWN.

    No. Every statement is tagged CONFIRMED (with its source), INFERRED (with the reasoning) or UNKNOWN (with what would settle it). Missing evidence appears as UNKNOWN, not as a guess.

    No. It uses roles and systems instead of names, avoids blame wording such as "failed to" or "human error", and asks what information a person lacked when they made a decision.

    No. It is a prompt-only skill that works from the text you give it. It runs no commands, makes no network connection and contacts no one.

    While it creates a comprehensive internal report, you can easily use the output to generate a customer-facing summary by asking the agent to strip internal specifics while maintaining the technical timeline.

    Updates are periodically released to refine the blameless logic and improve timeline accuracy; buyers will receive notifications and access to updated versions through the marketplace.

    New in DevOps & Cloud