SuperAPI AI Directory

product-methodology

Office & productivity

View repository

Updated 10-16 00:03

About this skill

Product Management Methodology

Canonical Output

The canonical proof of work for every engagement is an artifact pyramid — a three-layer progressively-disclosable structure at an absolute filesystem path. Every methodology and framework in this skill feeds into one of the three layers.

Respond to any caller with the absolute path to 00-index.md at the pyramid root. Not a summary. Not a handoff paragraph. A path.

The layer mapping is at references/artifact-pyramid-mapping.md.

Production Flow

Build the pyramid in this order. Each phase gates the next.

Phase 1: Scaffold

mkdir -p <pyramid-root>/{01-summary,02-analysis,03-dossiers}

Copy assets/pyramid-template.md to <pyramid-root>/00-index.md and fill in the overview.

Phase 2: Build L3 Dossiers

For each source (interview, competitive analysis, market data):

  1. Create a flat file directly in 03-dossiers/ — no subdirectories, no README files acting as folder indexes. A dossier is a single markdown file. If a topic has multiple sources, create multiple flat files (e.g., customer-interviews.md, competitive-analysis.md).
  2. Include source attribution metadata at the top:
    **Source:** URL or transcript identifier
    **Captured:** YYYY-MM-DD
    **Author:** Person or organization
    **Title:** Original document title
    
  3. Extract faithfully — no cherry-picking. Include counter-evidence.
  4. Add a NOTES section with any methodology context (how the data was collected, processed, etc.)

Gate C check: Every dossier has source metadata. Extracts are faithful. Methodology is documented.

Phase 3: Build L2 Analysis Files

From dossiers, compose one file per analysis dimension (problem, stories, scope, risk, etc.):

  1. Each file is self-contained — makes sense read alone

  2. Each file has a clear thesis, evidence, and conclusion

  3. Each file ends with a SOURCES section:

    SOURCES (LAYER 3 NAVIGATION)
    ../03-dossiers/customer-interviews/transcript-001.md
     -> Primary user pain evidence supporting Section 2
    
    ../03-dossiers/competitive-analysis/feature-matrix.md
     -> Competitor feature comparison referenced in Scope section
    

    Every file gets a SOURCES section. No exceptions. Even if the only L3 source is a stub ("Interviews pending — placeholder"), link to it. An orphaned file with no L3 traceability will be rejected at the gate.

Gate B check: Every claim traces to L3. Conflict transparency. Narrative structure. Interpretive value.

Phase 4: Build L1 Summary

From analysis files, compose the single L1 summary:

  1. Restate the research question / mission brief

  2. 3-5 key findings

  3. Implications for the audience

  4. End with a SOURCES section linking to every L2 analysis file:

    SOURCES (LAYER 2 NAVIGATION)
    ../02-analysis/01-problem-statement.md
     -> User pain this feature solves
    
    ../02-analysis/02-user-stories.md
     -> Canonical set of user needs
    

Gate A check: Every claim links to L2. Self-contained. Implications stated. No orphan claims.

Phase 5: Quality Gate Audit

Run this checklist before responding:

  • 00-index.md exists with navigation table
  • 01-summary/index.md has SOURCES linking to all L2 files
  • Every L2 file has SOURCES linking to L3 — no orphaned analysis files
  • Every L3 dossier has source attribution metadata
  • L3 dossiers are flat files in 03-dossiers/ — no subdirectories, no README placeholders
  • All SOURCES paths resolve correctly (relative from file location)
  • Each layer is independently consumable (a reader stops at any layer and has what they need)

Phase 6: Respond

Respond with ONLY the absolute path to 00-index.md. Example:

/tmp/pm-test-pyramid/00-index.md

No summary text, no analysis recap, no conversation. The caller reads the pyramid.

References

Reference Load when File
Artifact Pyramid mapping You need to map a specific methodology or artifact to its pyramid layer references/artifact-pyramid-mapping.md
RICE scoring You need to compare unrelated feature proposals by Reach × Impact × Confidence / Effort references/rice-framework.md
MoSCoW Scope is tight for a time-boxed release and you need crisp boundaries references/moscow-prioritization.md
Opportunity Solution Trees The problem space is messy and you need to connect customer needs to build decisions without jumping to solutions references/opportunity-solution-trees.md
Customer interview guide You're planning discovery interviews — how to structure them, what to ask, what to avoid references/customer-interview-guide.md
Spec template You need a requirements document that engineers, designers, and stakeholders can all work from references/spec-template.md
Stakeholder communication You're preparing a message for execs, engineers, designers, or customers — each has a different format references/stakeholder-communication.md
Decision log You've made a decision with tradeoffs that will be questioned later — log the context and expected outcome references/decision-log.md

Loading

Load a reference with:

skill_view(name="product-methodology", file_path="references/rice-framework.md")