agentsclimarketplace

Prompt iteration and diagnostics

Skill 0xhughs/director-skills/skills/prompt-iteration-and-diagnostics

Agent skills for AI filmmaking, cinematic prompting, and model-specific prompt export.

Install
npx -y skills add 0xhughs/director-skills --skill prompt-iteration-and-diagnostics

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Diagnose failed AI image or video outputs, identify root causes, revise prompts with single-variable iteration, and produce structured prompt revision reports.

SKILL.md

3.5 KB, 693 tokens by cl100k_base, as published. Nobody here has run it

prompt-iteration-and-diagnostics

When to use

  • The user shows or describes a failed output and asks how to fix it.
  • The user wants prompt diagnostics, iteration, A/B variants, failure analysis, or a revision report.
  • The output has drift, bad anatomy, wrong motion, weak realism, camera chaos, text/logo errors, prompt collapse, or model transfer failure.
  • The prose, dialogue, or film package has generic AI voice, weak story logic, rights/provenance gaps, or release-quality concerns.

When not to use

  • The user has no output or failure description and only needs first-draft prompting.
  • The issue is a tool outage or account/billing problem.
  • The user asks for unsafe bypass instructions.

Required inputs

  • Original prompt
  • Observed output or failure description
  • Target model/tool
  • Desired result

Optional inputs

  • Reference images
  • Settings/parameters
  • Seed/version
  • Previous attempts
  • Continuity bible

Workflow

  1. Restate the intended result and the actual failure.
  2. Classify failure: prompt ambiguity, contradiction, overload, model limitation, continuity drift, temporal overload, reference conflict, safety/policy rejection, or parameter mismatch.
  3. Identify the smallest change likely to improve the result.
  4. Apply the one-variable rule for iterative tests unless the prompt is fundamentally broken.
  5. Rewrite using the relevant skill: image, video, continuity, style, or model-adaptation.
  6. Produce a revision report with diagnosis, changed fields, unchanged anchors, expected improvement, and next test.
  7. When the failure is a model limitation, redesign the shot or route to a better model instead of forcing the same prompt.

Decision logic

  • If output is 80 percent correct, make minimal edits.
  • If identity drift appears, strengthen references and bible anchors.
  • If motion fails, reduce actions and camera moves.
  • If text/logos fail, choose a text-capable model or simplify typography.
  • If a prompt is rejected, remove unsafe/IP-sensitive content and do not provide bypass tactics.

Output formats

  • Failure diagnosis
  • Revised prompt
  • A/B test variants
  • Prompt revision report
  • Continuity update recommendations
  • Model-routing note
  • Story/voice/release QC report

Quality checks

  • Diagnosis maps to a specific prompt or model cause.
  • Revisions preserve what already worked.
  • Only meaningful variables change per test.
  • The report distinguishes fixable prompt issues from model limitations.
  • Story, voice, continuity, performance, rights, and provenance checks are separated when evaluating a full AI film package.
  • No unsafe bypass methods are included.

Anti-patterns

  • Rewriting the entire prompt after a small failure
  • Adding long negative lists without prioritization
  • Blaming the model before checking contradictions
  • Ignoring uploaded reference conflicts
  • Treating policy rejection as a prompt-engineering puzzle.

Exit criteria

  • The user has a revised prompt and a clear next test or model-routing decision.

Supporting files

Read only the supporting file needed for the active task:

  • references/iteration_diagnostics.md
  • references/failure_modes.md
  • references/revision_loops.md
  • references/story_voice_release_qc.md
  • templates/prompt_revision_report.md
  • templates/failed_output_intake.md

Gives 0 of the 12 instructions most prompt engineering skills give in 693 tokens

Counted across 563 of the 626 authors here whose files we hold, read 2026-08-06

  • ask at most three clarifying questionsin 22 of 563, across 15 files
  • respond in the user input languagein 14 of 563, across 9 files
  • preserve the original intentin 13 of 563, across 11 files
  • Establish baseline metrics and collect representative examplesin 12 of 563, across 2 files
  • Identify failure modes and prioritize high-impact fixesin 12 of 563, across 2 files
  • Apply prompt and workflow improvements with measurable goalsin 12 of 563, across 2 files
  • Roll back quickly if quality or safety metrics regressin 12 of 563, across 2 files
  • validate changes with tests and roll out in controlled stagesin 12 of 563, across 2 files
  • generate quantitative baseline performance reportsin 12 of 563, across 2 files
  • create representative test scenariosin 12 of 563, across 2 files
  • treat prompts as codein 12 of 563, across 5 files
  • test prompts on diverse inputsin 12 of 563, across 8 files

Said here and by no other author read

  • Restate the intended result and the actual failure
  • Identify the smallest change likely to improve the result
  • Change one variable per test unless the prompt is broken
  • Redesign the shot when the failure is a model limitation
  • Make minimal edits if the output is mostly correct
  • Strengthen references and anchors for identity drift

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.