Free Course: An AI Task Pipeline

We work with Jira through REST API v3 with Basic auth (email:token). Credentials live in a .env file in the project root: JIRA_BASE_URL, JIRA_EMAIL, JIRA_API_TOKEN, JIRA_PROJECT. Security rule: never print the API token to the chat, logs, code or commits — and never ask me to paste it into a message.

Step 0. If there is no .env file — create it with these four placeholder fields, check that .env is listed in .gitignore (add it if not), and tell me what to fill in where: the Jira address, the account email, the API token from id.atlassian.com and the project key. I will fill in the values myself. After that, stop and wait for my "done".

Once .env is filled in — verify the connection and scout my project:
1. GET {JIRA_BASE_URL}/rest/api/3/myself — show the account name from the response: this confirms the connection works.
2. GET /rest/api/3/project/{JIRA_PROJECT}/statuses — what issue types and statuses my board has.
3. Search issues by JQL (project = {JIRA_PROJECT} ORDER BY created DESC) — show the latest issues: key, title, status.
4. Pick any issue and request GET /rest/api/3/issue/{key}/transitions — which status transitions are available.

Finish with a summary: account, board statuses, available transitions and the list of issues in To Do. Do not change anything in Jira yet — this is recon only.

The Jira connection works — now lock the skill in so it applies in every new session. Create (or extend) the CLAUDE.md file in the project root and write this workflow into it:

## Working with Jira
- Credentials live in .env: JIRA_BASE_URL, JIRA_EMAIL, JIRA_API_TOKEN, JIRA_PROJECT. Never print the token to the chat, logs or code. Check that .env is in .gitignore — if not, add it.
- The command "pick up {issue key}" runs the full task cycle:
  1. GET /rest/api/3/issue/{key} — read the title, description and comments. If there is no acceptance criterion, formulate one and show it to me before starting.
  2. Move the issue to In Progress (POST /rest/api/3/issue/{key}/transitions).
  3. Do the work in the project code.
  4. Hand the result to a separate reviewer subagent: acceptance criterion match, bugs, edge cases, code style. Fix the findings immediately, before closing the task.
  5. Move the issue to Done (if the board has a Review or In Testing column — use it) and leave a comment: what was done, which files were touched.
- If the task is too big for one cycle — first propose splitting it into subtasks in Jira and wait for my confirmation.
- On the command "stop" — finish the current step cleanly and record the honest task status in Jira.

Write the workflow, show the final CLAUDE.md and confirm you are ready to work.

Pick up KAN-12.

Extend CLAUDE.md with an "AI review" section and from now on follow it when closing every task:

## AI review
- Before moving a task to Done, run a separate reviewer subagent. Hand it: the task's acceptance criterion and the full diff of the changes (git diff over the touched files).
- The reviewer's checklist: 1) acceptance criterion match; 2) bugs and edge cases — empty data, null, concurrent access; 3) security — secrets in the code, SQL injections, unprotected endpoints; 4) readability and project style.
- Every finding gets a priority: P0 — blocks closing the task, P1 — fix now, P2 — can be deferred (create a Jira task for it), P3 — a matter of taste, my call.
- Fix P0 and P1 immediately and re-run the review. A task may only be closed with an empty P0/P1 list.
- Post the review summary as a comment on the Jira issue: how many findings there were and which were fixed.

Write the section, show the final CLAUDE.md and confirm you are ready.

#!/bin/bash
# review.sh — AI review of a diff via the Anthropic API.
# Requires: curl, jq and a key:  export ANTHROPIC_API_KEY=sk-ant-...
# Usage:  ./review.sh [base_branch]

set -euo pipefail

BASE_BRANCH="${1:-main}"
DIFF=$(git diff "$BASE_BRANCH"...HEAD)

if [ -z "$DIFF" ]; then
    echo "No changes against $BASE_BRANCH — nothing to review."
    exit 0
fi

SYSTEM='You are a strict code reviewer. Check the diff against this checklist: 1) bugs and edge cases (empty data, null, concurrent access); 2) security — secrets in the code, SQL injections, unprotected endpoints; 3) readability and style. Every finding gets a priority: P0 (blocker), P1 (fix now), P2 (can be deferred), P3 (a matter of taste) — with the file and line. If there are no findings — reply APPROVED.'

curl -s https://api.anthropic.com/v1/messages \
    -H "x-api-key: $ANTHROPIC_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -H "content-type: application/json" \
    -d "$(jq -n --arg diff "$DIFF" --arg system "$SYSTEM" '{
        model: "claude-sonnet-5",
        max_tokens: 3000,
        system: $system,
        messages: [{role: "user", content: ("Review this diff:\n\n" + $diff)}]
    }')" | jq -r '.content[0].text'