Incident Postmortem Studio Pro

Turn incident records into an evidence-linked review your team can inspect, edit and follow through.

Incident Postmortem Studio Pro helps a capable agent reconstruct supplied incident records, retain uncertainty, assess explanations against evidence and prepare practical corrective-action follow-up. It works in English. It is a post-incident review workflow, not a production responder or an automatic root-cause detector.

A useful review, not just a polished template

Keep original UTF-8 source bytes, names and SHA256 hashes; link exact excerpts to stable event, assessment and action IDs. Preserve original timestamp literals and explicit offsets while ordering events in UTC. Choose metric endpoints deliberately: impact start to detection, detection to mitigation, and impact start to reviewed recovery. These are single-incident intervals, not mean-time metrics. Missing or unverified endpoints remain unavailable. Merge overlapping intervals only within the same named scope; do not invent affected-user counts or financial impact.

Separate observed mechanisms, changes, control gaps, hypotheses, counter evidence and unresolved explanations. A supported node does not prove its outgoing relationship; reviewed causal relationships need their own evidence and supported endpoints. Timing alone is not causation. Structural validation checks references and reviewer attestations, not source authenticity or semantic truth.

24 contexts and 96 complete fictional studies

Cover SaaS availability; API performance; databases; deployment/rollback; configuration/flags; DNS/certificates; dependencies; data pipelines; authentication; AI/agent workflows; checkout integrations; backup/restore; mobile offline sync; CDN/cache; container scheduling; storage/capacity; queues/events; monitoring/paging; batch/file delivery; webhooks; multi-region failover; security containment review; internal workflow automation; and research computing/HPC.

Each context has different trigger, mechanism and control-gap records, plus tailored controls and testable acceptance criteria. Four evidence-workflow variants per context cover reviewed replay, overlapping intervals, conflicting evidence and unresolved milestones. These are 96 original synthetic exercises, not 96 independent investigation methods or real customer incidents. Security and research examples do not establish a breach, regulatory compliance or scientific-result validity.

Edit the review without losing the source

An optional loopback workbench supports clean source import, exact excerpt selection, searchable evidence/reference pickers, guided event/assessment/link/action/narrative editing, event and excerpt search, draggable native diagrams and keyboard node movement. Use focus mode, undo/redo and project save/reopen. Unrelated accepted revisions preserve unsaved form drafts; apply each form before saving the accepted project. Deleting a referenced record is blocked and lists its dependents. Complex impact scopes/measures remain in the project/agent workflow. First user-source import from a fictional study clears fictional events, narrative and history.

Three actual HTML layouts—journal, command and briefing—combine with twelve visual foundations and internal/executive/customer reader views. Add a summary, what worked, what hindered response and learning notes. Draft narrative remains clearly marked and is excluded from customer views. Customer mode is an editorial filter, not anonymization or proof of certainty.

Follow-through with evidence

Set owners, due dates, priority and prevention/detection/mitigation/recovery/communication categories. Track HTTPS issue references and next-review notes; filter and sort actions by priority, owner and due date. A tracking reference does not create, fetch or synchronize a remote ticket. Reported done is separate from verified: verified closure requires an owner, acceptance criteria, verification evidence, named reviewer and actual review. Overdue uses the explicit review-as-of date. Review-next prompts expose gaps, not an approval score. Editable history records before/after values and source hashes; it is not tamper-proof attestation.

Genuinely editable generated work

The standard-library core creates self-contained HTML, native SVG timelines/assessment maps, CSV registers, project JSON, exact sources, private evidence/history and file manifests. Optional DOCX uses native text/tables; PPTX uses native text/shapes and attached connectors; XLSX uses six editable sheets and linked interval/status formulas. Action priority, categories, tracking and follow-up are retained in the native exports. Office files are editable snapshots: rebuild after source changes; external edits do not synchronize formats. Restrained Office layouts do not reproduce every HTML theme.

Requirements, delivery and validation boundaries

Run in the buyer's agent with permitted file and Python tools. No mandatory paid API, seller-hosted backend or remote log upload. The core uses Python 3.10+ standard library. Optional Word/PowerPoint require python-docx/python-pptx. Optional Excel uses host-provided Node.js and @oai/artifact-tool, or permitted host spreadsheet tools with unavailable formats disclosed. No silent global dependency installation. Direct source import supports UTF-8 text; retain binary originals separately and use permitted host extraction/OCR with limitations recorded.

Agensi loads instructions on demand. No independently installed or downloadable Skill package is promised. Generated user artifacts are distinct from the reusable Skill runtime; the helper does not copy its runtime/catalogue into the work pack. Agensi buyer auxiliary-file loading, temporary relative-file execution, optional dependencies and generated-work retrieval have not been verified in a real buyer account. Platform review is requested to verify these before approval. Microsoft Office desktop edit/save and independent real-incident customer acceptance also remain unverified. Local tests and previews do not prove those gates.

Creator-authored code and fictional fixtures include the MIT notice. Marketplace access/distribution remain subject to Agensi terms; user inputs retain their rights. Private packs contain original logs, excerpts and history. Share/customer reports retain narrative, IDs, source names and context: review access and wording before external use. Generation does not authorize production changes, notifications or publication.