- Home
- Skills
- Prompt & Skill Engineering
- Prompt Architect: Architect Mode
Prompt Architect: Architect Mode
Vague coding goal in, verified step plan out, with a ready-to-run, self-checking prompt for every step.
$19
Works with the AI Agents you already use
Prompt Architect: Architect Mode
Example session with this skill installed
Architect mode: add tags to blog posts, with filtering by tag. Target Claude Code.
- Read your context and instructions
- Compiled the prompt architect
Sample plan: verification commands below are proposed checks, not claims that they were run in this session.
Architect Plan: Add post tags and filter the Flask tutorial blog
As-of: 2026-11-05 · Size: Medium · Pattern: New feature · Target: Claude Code
Goal
Let authors add comma-separated tags to posts and let readers filter the blog index by tag.
Context
Facts (with sources)
- Flask uses SQLite through the standard-library sqlite3 module: flaskr/db.py.
- The schema drops and recreates tables: flaskr/schema.sql.
- Post create/update/index views are in flaskr/blog.py; tests use pytest and tests/data.sql.
- The tutorial app's test configuration is in pyproject.toml.
Assumptions (confirm if they differ)
- Tags are entered as one comma-separated field, trimmed and lowercased, with at most five tags of 1–20 characters. Impact: adjust the parser and its tests.
- This is the tutorial/local database, not a production database that must preserve data. Impact: if production data must be kept, add a migration plan before changing the schema.
Open questions
None for this bounded example; the assumptions are explicit and reversible.
Steps
| ID | Step | Depends on | Files | Verify by | Rollback |
|---|---|---|---|---|---|
| S1 | Add tag and post_tag tables plus test fixtures | none | flaskr/schema.sql, tests/data.sql, tests/test_db.py | pytest tests/test_db.py | revert changes |
| S2 | Add parsing and tag storage/query helpers | S1 | new flaskr/tags.py, tests/test_tags.py | pytest tests/test_tags.py | remove helper files |
| S3 | Save tags on post create/update | S2 | flaskr/blog.py, create/update templates, tests/test_blog.py | pytest tests/test_blog.py -k "create or update" | revert changes |
| S4 | Render tags and filter the index | S2 | flaskr/blog.py, index template, tests/test_blog.py | pytest tests/test_blog.py -k index | revert changes |
| S5 | Remove post-tag links on delete | S1 | flaskr/blog.py, tests/test_blog.py | pytest | revert changes |
Step prompts
S1: Add tag tables
Target
Claude Code with repo access.
Context
flaskr uses SQLite; flaskr/schema.sql defines its tables and tests/data.sql seeds test rows.
Task
Add tag and post_tag tables and a test fixture for a tagged post.
Constraints
Use primary/foreign keys; drop post_tag before its referenced tables; leave user and post definitions unchanged.
Output Format
Edit flaskr/schema.sql, tests/data.sql, and tests/test_db.py; add test_tag_tables_exist.
Acceptance Criteria
The new tables initialize and the fixture links a tag to a post.
Verification Checklist
- pytest tests/test_db.py passes.
- The test queries both new tables.
- Existing user/post definitions are unchanged.
S2: Add tag helpers
Target
Claude Code with repo access.
Context
The schema contains tag and post_tag tables; use the app's existing SQLite connection.
Task
Create parse_tags, set_post_tags, and get_tags_for_posts helpers and tests.
Constraints
Split commas, trim/lowercase, deduplicate, cap at five tags; use parameterized SQL and no new dependencies.
Output Format
Create flaskr/tags.py and tests/test_tags.py.
Acceptance Criteria
Helpers parse, replace, and fetch tags; bulk fetch uses one query.
Verification Checklist
- pytest tests/test_tags.py passes.
- Empty input returns an empty list/map as appropriate.
- set_post_tags does not commit the caller's transaction.
S3: Save tags on create/update
Target
Claude Code with repo access.
Context
flaskr/blog.py handles create/update and commits each post transaction; S2 supplies parse_tags and set_post_tags.
Task
Add a Tags field to both forms and save tags with the post.
Constraints
Use request.form.get('tags', ''); call the helper before the existing commit; preserve title validation and redirects.
Output Format
Edit the two forms, flaskr/blog.py, and tests/test_blog.py.
Acceptance Criteria
Create and update store the submitted tag links.
Verification Checklist
- pytest tests/test_blog.py -k "create or update" passes.
- Existing create/update tests still pass.
- Each view commits once.
S4: Show tags and filter index
Target
Claude Code with repo access.
Context
The index view renders posts; S2 supplies a bulk tag lookup.
Task
Render tags as links and filter by the ?tag= query parameter.
Constraints
Filter in SQL with parameterized values; unknown tags return an empty list, not a 404.
Output Format
Edit flaskr/blog.py, the index template, and tests/test_blog.py.
Acceptance Criteria
A tag link filters the list and an unknown tag shows no posts.
Verification Checklist
- pytest tests/test_blog.py -k index passes.
- The filter is parameterized SQL.
- The unfiltered index remains unchanged.
S5: Clean up links on delete
Target
Claude Code with repo access.
Context
SQLite foreign-key enforcement is off by default in this tutorial; deleting a post could leave post_tag rows.
Task
Delete the post_tag rows before deleting the post, in the same transaction.
Constraints
Do not change global foreign-key settings; tag-table cleanup is out of scope.
Output Format
Edit flaskr/blog.py and tests/test_blog.py.
Acceptance Criteria
Deleting a post leaves no post_tag rows for it.
Verification Checklist
- pytest passes.
- test_delete_removes_tags asserts no links remain.
- delete() commits once.
Self-review
B1–B8 plan checklist: pass. Prompt scores are heuristic checklist results, not guarantees of model behavior.
Connects securely to your tools. The creator never sees your data.
About this skill
Most agent failures on big tasks come from one fuzzy prompt trying to do six things at once. Architect Mode fixes that before any code gets written.
Give your agent a rough goal like "add tagging", "migrate auth to JWT", or "make /orders faster". It returns an Architect Plan:
- An insufficiency gate. If the goal is too vague to plan safely, it asks 3–5 sharp questions instead of guessing.
- Grounded context. It does a short, read-only scan of your repo and records facts with their sources. Anything it can't confirm goes under Assumptions, with the impact if the assumption is wrong. It never makes up file paths or APIs.
- A dependency-ordered step table. Each step has the files it touches, an observable Verify by check (a command, a test, or a visible behaviour), and a Rollback point.
- One copy-paste prompt per step. Each prompt follows Prompt Architect's 7-part contract: Target, Context, Task, Constraints, Output Format, Acceptance Criteria, Verification Checklist. Every prompt stands alone, so you can hand step 3 to Codex, a teammate, or a fresh chat.
- Self-review before handoff. An 8-point plan rubric and a 4-dimension prompt score (clarity, specificity, structure, actionability). Anything under 80 gets rewritten once before you see it.
- Act mode, when you ask for it. "Execute S1" runs one step at a time. It verifies each step, checkpoints, and stops on the first failure.
Also included:
- Refine mode: tighten any existing prompt and get a "Changes made" list.
- Prompt-only mode: one great prompt for a single task.
- Eight goal-pattern playbooks: feature, refactor, migration, bug hunt, performance, test coverage, integration, upgrade.
- Per-model adjustments for Claude, GPT/Codex, Gemini, Cursor, and human readers.
- An optional
prompt_lint.py. It checks plans and prompts against the contract: dependency cycles, missing checks, placeholders, vague phrases, and leaked secrets. It is standard-library Python and makes no network calls.
This skill is built from the prompt program behind Prompt Architect (Meraby Labs), rebuilt to run inside any coding agent with no IDE extension, no backend, and no API keys.
Tested: On the public Flask tutorial app, two of the generated step prompts were run exactly as written. The test suite went from 24 to 29 passing, and every verification checklist held. Both sample plans lint clean.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
A detailed milestone may include up to 15 steps; Epic goals are split into milestones, with later milestones outlined rather than fully prompted. Repo-specific fact-gathering requires a repo or user-provided context; otherwise assumptions are labeled and the skill may ask clarifying questions. Act mode requires an explicit request and works one step at a time; execution depends on the agent’s tools and permissions.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 11 days ago
- Software
Git
- Optional
Python 3.8+
Python is only for the linter.
Permissions
No network calls or credentials are required. The package includes one optional Python 3.8+ standard-library linter (19,139 bytes, under 20 KB); no third-party dependencies.