fetch

fetch

Prepare selected Prism instruments as an instruction block that a subagent can consume directly.

Role

If /prism search is "discover what exists," fetch is "get the selected ones ready."

• Prepares paths and read directives so the subagent can actually open and follow the instrument procedures.
• Does not inline instrument content — returns absolute paths + read instructions + brief summary only.
• Typical workflow: /prism search to find candidates → fetch to equip selected instruments.

Input

Accepts one or more of the following:

• Instrument name (slug), one or more — stride, owasp-top10, systems-thinking, etc. Must match a name value in catalog.yml.
• Domain + class combination — "all security lenses", "all education frames" — fetches every matching instrument.
• If a name is ambiguous, show candidates from catalog.yml and ask one clarifying question. Do not guess.

Resolve algorithm

3-layer path lookup

Prism stores instruments in three layers. Merge with precedence project > global > bundle:

1. Bundle layer (read-only) — instruments shipped with the plugin.

   • Catalog: <plugin-root>/catalog.yml
   • Library: <plugin-root>/library/

2. Global layer (optional) — user-wide instruments.

   • Catalog: ~/.claude/prism/catalog.yml
   • Library: ~/.claude/prism/library/
   • Silently skip if absent.

3. Project layer (optional) — project-scoped instruments.

   • Catalog: ./.claude/prism/catalog.yml (relative to CWD)
   • Library: ./.claude/prism/library/
   • Silently skip if absent.

Merge and path resolution

• Read catalog.yml from all three layers and extract the path field for each requested instrument.
• On name collision, higher-precedence layer wins. Note overrides in output.
• Convert each path to an absolute path relative to its layer root.
• Verify existence (Glob or Read). If missing, emit a warning and skip that entry.
• Extract one_liner and class from frontmatter as the brief summary.

Output format

Return a copy-pasteable text block for subagent prompts:

## Prism Instruments — Subagent Directive

Read the following instrument files with the Read tool and follow
each file's core procedure section in your analysis workflow.

| # | name | class | summary | path |
|---|------|-------|---------|------|
| 1 | stride | lens | Identify threats per STRIDE category | /full/path/to/stride.md |
| 2 | owasp-top10 | lens | Run OWASP Top 10 checklist | /full/path/to/owasp-top10.md |

### Core section guide by class
- **lens**: Follow the "Analytical Procedure" steps in order
- **frame**: Apply "Categories" + "Classification Procedure"
- **model**: Apply "Core Variables" + "Application Procedure"
- **stance**: Set analysis perspective via "Guiding Questions"
- **heuristic**: Use "The Rule" + "When It Applies" as checkpoints

If overrides occurred, append a note below the table:

notes:
  - 'stride' in project layer overrides bundle entry (layer=project used)

Operational rules

• READ-ONLY: never create or modify files. Read catalog and library only.
• Re-read catalog.yml every invocation — do not cache.
• If more than 10 instruments are requested, emit a warning first and proceed only after user confirmation.
• Delivery is path-based: subagents read files themselves via the Read tool. Do not inline instrument bodies.

NOT this skill

• Catalog browsing/search — use /prism search.
• Instrument creation — use /prism <framework>.
• Agent/skill creation — use Claude Code's native /agents flow. Prism supplies the catalog, not agent configs.