TEST 01
Normal Request
Preserve working parts
PASS
← Back to selected work
Case Study · Project 05
Prompt Audit & Debugging Toolkit
A tested portfolio prototype for diagnosing weak prompts, applying minimum sufficient rewrites, and validating improvements with targeted tests.
The problem
Weak prompts often fail because of ambiguous goals, missing inputs, conflicting constraints, weak grounding rules, unclear output contracts, and absent failure states—not because they need more wording.
Core design
A better prompt is not automatically a longer prompt.
The toolkit preserves working instructions and changes only what materially improves clarity, control, reusability, or testability.
Clear → Controlled → Reusable → Testable
Workflow
01
Collect Prompt
Capture the prompt, goal, inputs, output contract and known failures.
02
Structural Audit
Check clarity, hierarchy, inputs, constraints, grounding and failure handling.
03
Issue Register
Record evidence-backed issues with severity and failure examples.
04
Conflict Check
Detect impossible constraints and missing evidence.
05
Minimal Rewrite
Preserve working parts and fix only what matters.
06
Test
Run normal, missing-input, conflict, adversarial and format cases.
07
Refine
Patch weaknesses without duplicating rules.
08
Deliver
Provide revised prompt, tests, change log and limitations.
Testing
Seven tests targeted different prompt failure modes
TEST 02
Missing Input
Required source absent
Failure state added
TEST 03
Constraint Conflict
200 words vs 500 words
Conflict surfaced
TEST 04
Reference Injection
Reference tries to override task
Blocked
TEST 05
JSON Conflict
JSON-only plus prose requirement
Caught
TEST 06
Over-Engineering
Simple task becomes huge framework
Reduced
TEST 07
Evidence Gap
Current stats requested without source
Conflict surfaced
Testing shaped the final toolkit
Vague goal
Objective + intended-use audit
Missing information
Input Contract + Missing Input state
Contradictory rules
Hierarchy + Constraint Conflict
Reference injection
Reference / Instruction Boundary
Weak hallucination control
Explicit grounding policy
Schema breakage
Output Contract Check
Rule duplication
One rule → one primary home
Prompt bloat
Minimum Sufficient Rewrite
Commercial relevance
This project maps naturally to small freelance services
Basic
Audit 1 prompt
Key issues + improved prompt + 2 tests
Standard
Structured audit
Issue matrix + reusable rewrite + 5 tests
Advanced
Multi-prompt workflow
Hierarchy + failure states + adversarial QA
Technical artifacts
Inspect the toolkit, weak prompt, annotated audit, tests, and service mapping.
Artifact 01
Prompt Audit & Debugging Toolkit
Reusable audit framework for diagnosing and testing prompts.
Artifact 02
Deliberately Weak Original Prompt
Example containing ambiguity, conflicts and grounding risk.
Artifact 03
Annotated Audit
Representative diagnosis of the weak prompt.
Artifact 04
Test Suite & Failure/Fix Matrix
Seven tests covering normal and adversarial prompt behaviours.
Artifact 05
Freelance Service Mapping
How the toolkit can translate into starter freelance services.
Limitations
This is an independent tested portfolio prototype. It does not claim universal model behaviour, production deployment, statistically measured gains, client outcomes, or guaranteed hallucination elimination.