AI Blitz — devops-lib task catalogue
Tasks identified for the AI Blitz Week 3 capstone (Task 11 from the Week 3 playbook). Each task is a small, agent-doable unit of work grounded in real findings from BUGS_AND_IMPROVEMENTS_REPORT.md, docs/tribal-knowledge.md, and review-learnings.md.
Mix
5 tasks: 3 small features + 2 bug fixes.
| # | Task | Type | Files touched (rough) | Difficulty |
|---|---|---|---|---|
| 01 | task-01-securityScan-hardcoded-ip.md |
Bug fix (P0 security) | 2 | low |
| 02 | task-02-checkoutSubmodule-typo.md |
Bug fix (correctness) | 1-2 + consumer repos | medium |
| 03 | task-03-onlyPushtoJfrog-jdk-versions.md |
Small feature | 1 | low–medium |
| 04 | task-04-buildPipeline-agent-label-param.md |
Small feature | 3 | low |
| 05 | task-05-cloudFunctionCICD-implementation.md |
Small feature | 4-5 + 1 consumer | high |
Capstone picks
| Day | Mode | Task | Why this pick |
|---|---|---|---|
| Day 4 (Thu) | Interactive (war-room) | Task 05 — Implement cloudFunctionCICD |
Real design decisions throughout (gen-1 vs gen-2 syntax, runtime → buildObjHelper mapping, service-account auth path). Maximises learning signal when humans correct the agent in real time. |
| Day 5 (Fri) | Autonomous | Task 02 — Fix chekoutSubmodule typo |
Clear acceptance criteria, mostly mechanical, the deprecated-alias-with-log-warning pattern is recognisable. Agent can solo without supervision. |
Adding a new task
- Copy
_TEMPLATE.mdtotask-NN-<short-slug>.md. - Number sequentially (next is 06).
- Fill out Goal + Acceptance Criteria. Cite a source —
BUGS_AND_IMPROVEMENTS_REPORT.md§,docs/tribal-knowledge.md§, or a specificreview-learnings.mdrule ID. - Update the table above.
Source documents the catalogue draws from
../BUGS_AND_IMPROVEMENTS_REPORT.md— known bugs + tech debt../docs/tribal-knowledge.md— non-obvious conventions../review-learnings.md— PR-review-derived rules../docs/architecture.md— module map, downstream services, invariants
Out of scope
Tasks not in this catalogue (good candidates, but not picked):
- Full test-harness bootstrap (JenkinsPipelineUnit) + first unit tests — large scaffolding effort, deferred to a dedicated PR.
- ArgoCD
--password-stdinrefactor — pure refactor, no behavioural change; lower learning value. - Refactor
validate_configs.py(1207 lines) into modules — too large for a single Blitz task. xq(XML query) → Groovy parser on Gradle files inbuildGradle.groovy— same reason.