More screenshots
Works with the AI Agents you already use
Engineering Workflow Skills: 14 Senior-Engineer Workflows
Fourteen checkable workflows, from clarifying a request to shipping with a rollback, that make an AI coding agent work like a careful senior engineer.
$9.99
Engineering Workflow Skills: 14 Senior-Engineer Workflows
Example session with this skill installed
test_summary_lists_categories passes on my machine and fails about half the time in CI. Find out why.
- Read your context and instructions
- Compiled the engineering workflow
Symptom. Observed: assert summarize(ORDER) == "3 items in: books, games" fails with '3 items in: games, books'. Expected: "3 items in: books, games".
Reproduction. Failure rate: 13 of 20 runs. Each run is a new process. A fixed hash seed makes it repeatable: PYTHONHASHSEED=2 failed, PYTHONHASHSEED=0 passed. Seeds 0, 1 and 4 passed. Seeds 2, 3 and 5 failed.
Hypotheses.
| # | Hypothesis | Prediction | Result |
|---|---|---|---|
| H1 | Output depends on set iteration order | Same input prints a different order under different PYTHONHASHSEED values | Confirmed: seed 2 printed games, books, seed 0 printed books, games |
| H2 | The test depends on test order or shared state | The failure would not follow the hash seed | Refuted: the failure follows the seed and the test runs alone |
| H3 | Locale or encoding changes the sort | Failure would change with LC_ALL | Not tested: H1 explained every observation |
Root cause. Because summarize builds tags as a set and joins it directly, when Python randomises string hashing in each new process, the categories come out in a different order; the fix belongs in orders/summary.py:6, where the join happens. A prediction made from this sentence was checked: with PYTHONHASHSEED=2 the set prints as {'games', 'books'}.
Fix and regression test.
- Fix,
orders/summary.py:6:', '.join(tags)became', '.join(sorted(tags)). - Regression test,
tests/test_summary_order.py: runssummarizein a new process for hash seeds 0 to 5. On the parent commit: 3 failed, 3 passed. With the fix: all pass. - Original reproduction after the fix: 0 failures in 20 runs. Full suite:
7 passed.
Sibling search. git grep -n "join(" -- '*.py' found one match, the line just fixed. No other place joins an unordered collection.
Commit message.
Fix flaky category order in order summary
Cause: summarize() joined a set, whose order changes with Python's per-process hash seed.
Evidence: PYTHONHASHSEED=2 prints "games, books"; seeds 2, 3 and 5 failed, 0, 1 and 4 passed.
Test: tests/test_summary_order.py runs seeds 0 to 5; it fails 3 of 6 before the fix and passes after.
Connects securely to your tools. The creator never sees your data.
About this skill
AI coding agents are fast. They also skip steps: they guess at requirements, fix symptoms, weaken a test to make it pass, and ship with no way back. This pack gives your agent the workflow a careful senior engineer follows for the fourteen jobs that matter most.
How it works. You ask for the job in your own words ("debug this failing test", "plan this migration", "review my PR") and the agent picks the matching skill from its description. Every skill has the same shape, so results are consistent: numbered steps, a defined output, exit criteria the agent can check with a command or a number, the ways that job usually goes wrong, and the points where the agent must stop and ask you.
The fourteen skills. Clarify requirements, write a feature spec, break down tasks, design an HTTP API, plan a database migration, make a test-first change, refactor without behaviour change, debug from evidence, investigate performance, give and receive code review, review the security surface, release with rollback, write an incident postmortem, and write technical docs.
What makes it different. Exit criteria are checkable ("the reproduction passes 10 of 10 runs"), never "looks good". Each skill says when not to use it, so the right one triggers. Each defines its output, so you get the same report layout every time.
Where it stops. These are process skills. They do not know your stack's conventions, so the agent still reads your code first. They do not replace human approval for releases, migrations or security decisions, and each skill says when the agent must hand the decision to you.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
- Process guidance, not code generation: the agent still does the engineering work.
- Written for general software teams. Your own conventions win where they differ.
- Skills ask the agent to run commands your project already has; they cannot supply tools you lack.
- No skill replaces human sign-off on releases, migrations or security findings.
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 today
- Software
Git
Permissions
File Scopes
None of its own. When a skill runs, the agent may run git, the project's test runner and its linters. No network access, no credentials.