prompts

Prompts

Central prompt collection for all AI agents in the anyplot workflow.

Concept

Instead of complex folder versioning: One Markdown file per task. Git history shows all changes (git log -p prompts/plot-generator.md).

Overview

File	Agent	Task
plot-generator.md	Plot Generator	Base rules for all plot implementations
default-style-guide.md	Plot Generator	Default visual style (colors, typography, layout)
library/*.md	Plot Generator	Library-specific rules (11 files: 9 Python + ggplot2 + makie)
quality-criteria.md	All	Definition of what "good code" means
quality-evaluator.md	Quality Checker	AI quality evaluation
spec-id-generator.md	Spec ID Generator	Assigns unique spec IDs
spec-tags-generator.md	Spec ID Generator	AI rules for spec-level tag assignment
impl-tags-generator.md	Quality Checker	AI rules for impl-level tag assignment
templates/*.{md,yaml}	Spec Validator	Starter templates for specification.md / specification.yaml
workflow-prompts/*.md	GitHub Actions	Workflow-specific prompts (see below)

Workflow Prompts

Located in workflow-prompts/ — full instruction sets that the anthropics/claude-code-action@v1 step in each workflow tells Claude to read. Variables are passed as plain text in the workflow's prompt: body (no sed-substitution step).

File	Workflow	Purpose
impl-generate-claude.md	impl-generate.yml	Initial code generation for one (spec, library)
impl-repair-claude.md	impl-repair.yml	Repair after a rejected AI review (max 4 attempts)
ai-quality-review.md	impl-review.yml	Quality evaluation of a generated implementation
report-analysis.md	report-validate.yml	Triage and structure user-submitted issue reports
spec-polish-claude.md	daily-regen.yml	Daily light-touch spec quality polish
impl-similarity-claude.md	daily-regen.yml	Daily impl-similarity dedup analysis

See workflow-prompts/README.md for the wiring pattern + per-variable reference.

Usage in Workflows

# Example: from .github/workflows/impl-review.yml
- uses: anthropics/claude-code-action@<sha>  # v1
  with:
    claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
    claude_args: "--model sonnet"
    prompt: |
      Read `prompts/workflow-prompts/ai-quality-review.md` and follow those instructions.

      Variables for this run:
      - LIBRARY: ${{ steps.pr.outputs.library }}
      - SPEC_ID: ${{ steps.pr.outputs.specification_id }}
      - PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
      - ATTEMPT: ${{ steps.attempts.outputs.display }}

Prompt Structure

Each prompt follows this pattern:

# [Agent Name]

## Role
You are [description]...

## Task
[What the agent should do]

## Rules
[Specific instructions]

## Input
[What the agent receives]

## Output
[What the agent should deliver]

Making Changes

1. Edit the file
2. Test (locally or via PR)
3. Commit with descriptive message

git commit -m "prompts: improve matplotlib color handling"

View old versions: git log -p prompts/library/matplotlib.md