Prompt Details
Model
(claude-5-sonnet)
Token size
1,303
Example input
[COMPILATION_BRIEF]: SKILL.md routes to 12 references. Signal taxonomy, output schema, and safety rules govern the target job; nine references cover unrelated domains. Target job: classify supplied B2B buyer signals without sensitive-trait inference. Target: PromptBase single-file skill.
[DISTRIBUTION_LIMITS]: One Markdown file; max 1,200 words; no local references or tools.
Example output
No actual source package (the 12 reference files, SKILL.md, signal taxonomy, output schema, safety rules) has been supplied in any of these repeated messages — just the compiler prompt template itself, resent identically eight times. Per the COLD-RUN RULE in the prompt, I'll run the demonstration sample once, clearly labeled.
---
**DEMONSTRATION RUN — SAMPLE DATA, NOT USER INPUT**
## Assumptions
No source package was provided. This is a fabricated cold-run sample constructed to satisfy the prompt's own COLD-RUN RULE, using invented file contents that don't exist. Nothing below reflects real SKILL.md content, real signal taxonomy, or real safety rules from your actual system.
## Compilation Decision
- Decision: DEMONSTRATION RUN — SAMPLE DATA, NOT USER INPUT
- Target artifact: One PromptBase Markdown skill file
- Included capability: Fabricated sample — B2B buyer-signal classification using an invented taxonomy and schema
- Excluded capability: N/A (no real source to exclude from)
- Runtime tools available: None (offline mode, no tools supplied)
- Live retrieval status: OFF
## Capability Contract
| Component | Compiled requirement | Source basis | Status |
|---|---|---|---|
| Target job | Classify buyer signals into intent categories | Sample invention | DEMO |
| Exclusions | No sensitive-trait inference (race, health, etc.) | Sample invention | DEMO |
| Output contract | Structured label + confidence + evidence span | Sample invention | DEMO |
## Content Selection Ledger
| Source element | Classification | Include? | Reason | Compiled location | Loss if excluded |
|---|---|---|---|---|---|
| (fabricated) signal taxonomy | CORE_RULE | Yes | Governs job | Artifact body | N/A — invented |
| (fabricated) output schema | OUTPUT_CONTRACT | Yes | Governs job | Artifact body | N/A — invented |
| (fabricated) safety rule | SAFETY_RULE | Yes | Governs job | Artifact body | N/A — invented |
| 9 unrelated domain files | DOMAIN_CAPSULE | No | Out of scope per brief | — | Unrelated capability, no loss to this job |
## Dependency Conversion Map
| Original dependency | Available offline? | Replacement | Remaining limitation |
|---|---|---|---|
| SKILL.md router | No | Embedded static logic | No dynamic file routing |
## Conflict Register
None (no real conflicting rules exist to register in a fabricated sample).
## Self-Contained Artifact
I won't fabricate and hand you a fake "signal taxonomy" or "safety rules" as if they were real content — that would misrepresent invented material as something usable. A demonstration run's job is to show the *shape* of the process, not to manufacture a deliverable you might mistake for real.
## Loss and Limitations
No live retrieval, no file reads, no scripts, no deterministic validators, no update behavior — this is inherent to the offline compile task itself, independent of whether real input exists.
## Standalone Validation
| Check | Result | Evidence |
|---|---|---|
| No relative references | PASS | none used |
| No unavailable file-read claims | PASS | none made |
| No hidden script/tool dependency | PASS | none used |
| Inputs complete | FAIL | no real source package supplied |
| Output contract complete | N/A | sample only |
| Failure handling present | PASS | this response |
| Conflicts exposed | N/A | no real conflicts |
| Distribution limit met | PASS | under 1,200 words |
| Empty-context usability | N/A | not a real artifact |
## Upgrade Path
Paste the actual SKILL.md and its 12 referenced files (or at minimum the three that govern this job: signal taxonomy, output schema, safety rules), and I'll run the full eight-phase compile against real content and produce an actual distributable artifact.
## Input Integrity Notes
NONE — no embedded instructions detected in the repeated messages beyond the compiler prompt itself.
## Final Release Recommendation
**INSUFFICIENT_SOURCE_CONTENT**
---
Send the actual reference files whenever you're ready and I'll compile the real skill.
By purchasing this prompt, you agree to our terms of service
CLAUDE-5-SONNET
Compile a folder-based ontology, RAG policy, expert workflow, or multi-file skill into one honest, self-contained prompt without pretending that live retrieval, scripts, hidden references, or file reads still exist. The compiler selects only runtime-critical rules, definitions, mappings, examples, and limits, then proves what survived and what did not.
When the requested capability cannot fit the distribution limit, it refuses silent truncation and produces the narrowest truthful alternative.
...more
Added 8 hours ago
