SuperAPI AI Directory

agent-progress-visualizer

Office & productivity

View repository

Updated 10-16 06:52

About this skill

Agent Progress Visualizer

This skill creates and maintains a user-facing progress cockpit for UI/UX Agent Skill System projects. It translates canonical task-plan state into plain-language HTML for non-technical users.

It does not execute project tasks and does not own canonical status. Before planning, project-request.json and workflow-log.jsonl are only bootstrap sources. After planning, TASK-PLAN.md becomes the source of truth.

Operating Modes

  • bootstrap: create the first visible progress screen immediately after the system accepts a substantial user request, before TASK-PLAN.md exists.
  • init: create or replace the first task-plan-derived progress-state and progress screen after TASK-PLAN.md exists.
  • update: regenerate state and HTML after task-plan, report, artifact, feedback, or wiki changes.
  • review: inspect an existing progress screen for source-of-truth drift, fake status, privacy leaks, or confusing copy.
  • feedback-gate: expose user decisions and comments, then hand them to design-feedback-collector.
  • developer-view: reveal technical task ids, skill routes, dependencies, and artifact paths without changing customer copy.

Required Workflow

  1. Read references/data-source-contract.md.
  2. Read references/decision-framework.md.
  3. Read references/safety-boundaries.md.
  4. In bootstrap mode, do not wait for TASK-PLAN.md. Immediately create project-request.json, append workflow-log.jsonl, build a minimal progress-state.json, render agent-progress-screen.html, and open it in the browser. In the local project implementation, use bootstrap_progress_screen.py --open-browser.
  5. After the bootstrap screen exists, route to task-plan-v2-orchestrator to create canonical TASK-PLAN.md.
  6. In init or update mode, locate canonical TASK-PLAN.md. If no canonical task plan exists and bootstrap has not run, run bootstrap first, then route to task-plan-v2-orchestrator.
  7. Build or update progress-state.json from canonical task-plan fields, reports, artifacts, feedback state, and sanitized wiki decisions.
  8. If a declared journey exists, include journey_alignment from the journey registry and route skip/rework/block decisions through workflow-compliance-supervisor.
  9. Render agent-progress-screen.html as a derived view. Do not edit HTML as the only state change.
  10. In task-plan-derived init mode, open the rendered HTML in the browser immediately after creation. In the local project implementation, pass --open-browser to render_progress_screen.py. In update mode, open it only when the user asks or is actively reviewing.
  11. For user choices, comments, approvals, or rejections, route durable capture to design-feedback-collector.
  12. For large multi-session projects, route accepted decisions to project-wiki-manager after sanitization.
  13. Validate consistency before claiming the progress screen is ready.

Source-Of-Truth Rules

  • In bootstrap mode, project-request.json and workflow-log.jsonl are temporary startup sources.
  • After planning, TASK-PLAN.md is canonical.
  • progress-state.json is derived machine-readable state.
  • agent-progress-screen.html is a derived visual layer.
  • If Markdown, JSON, and HTML disagree, fix Markdown or the generator first, then regenerate.
  • Browser-only localStorage or clipboard state is not accepted as the only record of a decision.

What To Generate

Default local artifacts:

project-request.json
TASK-PLAN.md
progress-state.json
agent-progress-screen.html
workflow-log.jsonl
feedback/progress-feedback-register.schema.json
reports/progress-validation-summary.txt

Use assets/progress-state.template.json when a project needs a starter state before a generator exists.

Human-Language Rules

  • Customer mode says "Collecting search queries", not serp-keyword-harvester.
  • Customer mode says "Waiting for your selection", not required_approvals.
  • Developer mode may show task ids, skill routes, dependencies, and artifact paths.
  • Never show secrets, tokens, recovery codes, cookies, .env values, private screenshots, or private URLs.

Validation

When using the local project implementation, prefer these checks:

python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/bootstrap_progress_screen.py" --project-root "$PROJECT_ROOT" --project-title "$PROJECT_TITLE" --user-request "$USER_REQUEST" --journey-registry "$UIUX_SKILL_SYSTEM_ROOT/data/journey-registry.json" --declared-journey new-premium-site --open-browser
python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/check_bootstrap_progress.py" --project-root "$PROJECT_ROOT"
python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/generate_progress_state.py" --task-plan "$UIUX_SKILL_SYSTEM_ROOT/TASK-PLAN.md" --out "$UIUX_SKILL_SYSTEM_ROOT/progress-state.json"
python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/render_progress_screen.py" --state "$UIUX_SKILL_SYSTEM_ROOT/progress-state.json" --out "$UIUX_SKILL_SYSTEM_ROOT/agent-progress-screen.html"
python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/check_progress_consistency.py" --task-plan "$UIUX_SKILL_SYSTEM_ROOT/TASK-PLAN.md" --state "$UIUX_SKILL_SYSTEM_ROOT/progress-state.json" --html "$UIUX_SKILL_SYSTEM_ROOT/agent-progress-screen.html"
python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/validate_workflow_compliance.py" --fixtures
python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/no_secret_scan.py" "$UIUX_SKILL_SYSTEM_ROOT"

For first creation of a user-facing project progress screen before TASK-PLAN.md, bootstrap with browser opening:

python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/bootstrap_progress_screen.py" --project-root "$PROJECT_ROOT" --project-title "$PROJECT_TITLE" --user-request "$USER_REQUEST" --open-browser

After TASK-PLAN.md exists, replace the bootstrap state with task-plan-derived state:

python3 "$UIUX_SKILL_SYSTEM_ROOT/scripts/render_progress_screen.py" --state "$PROJECT_ROOT/progress-state.json" --out "$PROJECT_ROOT/agent-progress-screen.html" --open-browser

Skill package checks:

python3 $CODEX_HOME/skills/.system/skill-creator/scripts/quick_validate.py $CODEX_HOME/skills/agent-progress-visualizer
python3 $CODEX_HOME/skills/senior-skill-architect/scripts/lint_production_skill.py $CODEX_HOME/skills/agent-progress-visualizer

Handoff Shape

Progress mode:
Bootstrap request:
Canonical task plan:
Generated state:
Generated HTML:
User gates:
Feedback artifact:
Wiki sync:
Checks run:
Skipped checks:
Next owner: