CtrlK
BlogDocsLog inGet started
Tessl Logo

jbaruch/context-workshop

Decomposed ticket workflow from the AI Native DevCon London 2026 context workshop: skills for code and documentation tickets, the mandatory tests-first / clean-review / human-review rules, and the deterministic scripts/ they lean on.

70

Quality

88%

Does it follow best practices?

Impact

—

No eval scenarios have been run

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

SKILL.mdskills/code-ticket/

name:
code-ticket
description:
Workflow for a ticket that touches real source code (anything that is not purely documentation). Use when a ticket changes code, or when you cannot tell whether it is docs or code (default to this). Covers tests-first gating, implementation, running the full suite, PR, Copilot review, and merge-on-clean-review.

Code ticket workflow

Use this when a ticket touches any real source code (anything that is not purely documentation). If you genuinely cannot tell whether a ticket is documentation or code, treat it as a code ticket — code changes are the riskier ones, so it is always safer to take the more careful path.

The deterministic mechanics live in scripts/ (run them from the repo root). This skill is the judgment: deciding what to build and how to fix failures and review comments. Two mandatory gates apply — no tests, no PR and merge only on a clean review; follow them exactly.

Steps

  1. Tests-first gate (rule). Run python3 scripts/run_tests.py --check. If it exits non-zero, stop here — write tests first before implementing or opening a PR.
  2. Implement what the ticket asked for.
  3. Run the whole suite with python3 scripts/run_tests.py. If it exits non-zero, read the failures, fix the code, and run it again. Repeat until it exits 0 — do not move on while even one test fails.
  4. Open the PR: python3 scripts/open_pr.py --title "<title>" --body "<body>". You supply the title and body (judgment); the script creates the branch if needed, commits, pushes, and opens the PR against main. It prints the PR URL.
  5. Request a Copilot review: python3 scripts/request_copilot_review.py <pr-number>. (The script resolves the PR and reviewer ids and fires the request — no manual GraphQL.)
  6. Wait for the review: python3 scripts/await_copilot_review.py <pr-number>. It blocks until Copilot has reviewed, then prints JSON: {"state", "clean", "comments"}.
  7. If clean is false, address every comment in the JSON: fix the code, push, then re-run step 5 and step 6. Repeat this fix-summon-wait loop as many times as needed.
  8. Merge gate (rule). Only proceed once the review JSON reports "clean": true.
  9. When the review is clean, merge: python3 scripts/merge_pr.py <pr-number>.

skills

code-ticket

SKILL.md

README.md

tile.json