AI Development Methodology

tinker — example adopter project

Fictional. Generated as a worked example of the AI Development Methodology applied to a project from scratch. Not a real product.

What tinker is

A small command-line utility for capturing developer notes during coding sessions. Notes are associated with a working directory + a calendar event (when one is in progress). Built for solo developers who want a low-friction way to log context that would otherwise be lost.

Target user: an individual developer; not a team product.

Distribution: open-source CLI on package managers; one-line install.

Business model: free, MIT-licensed; no paid tier in v1.

How tinker adopts the methodology

The artifacts under this folder match the methodology’s prescribed shape:

Methodology layer	tinker’s instance
Strategy master plan	strategy/00_master_plan.md
Pillars	pillars/P1_capture.md, pillars/P2_retrieval.md
Epic rollup	backlog/EPICS.md
First epic charter	backlog/epics/E01-cli-foundations/README.md
Active items in that epic	backlog/epics/E01-cli-foundations/BACKLOG.md (4 items)
Archived items	backlog/epics/E01-cli-foundations/ARCHIVE.md (1 done item — BL-0001)
Deferred (P3) items	backlog/epics/E01-cli-foundations/FUTURE.md (3 items using Scheme B IDs BL-E01-F##)
Epic acceptance + regression tests	backlog/epics/E01-cli-foundations/TEST.md
Cross-epic manual-QA queue	backlog/TEST_BACKLOG.md
Work blocked on human agency	backlog/HUMAN_NEEDED.md (1 active — the storage-backend decision blocking BL-0005)
Volatile session state	backlog/ACTIVE_CONTEXT.md (mid-item on BL-0002)
Lessons-learned memory	memory/MEMORY.md + 3 active entries + 1 archived

This is the minimum-viable set for an adopter to see the methodology operationalized. A real adopter project would add more pillars, more epics, more items as the project grew; this example shows the starting shape.

Conventions worth looking at specifically, because they are easier to copy than to describe:

• Code Maps on the two Effort-M items (BL-0002, BL-0005) — annotated paths, the utility to reuse, and the constraint that isn’t obvious from the code. The S/XS items keep a plain Files (probable): list, which is the point: the upgrade is keyed to effort, not applied everywhere.
• Frozen intent markers on BL-0002 and on the E01 charter — the approved goal and criteria are human-owned; everything else in the body stays editable.
• The blocked-item protocol, end to end. BL-0005 is blocked with Lock: — released, a Blocker: line in the body, and a matching entry in HUMAN_NEEDED.md — so the agent moved on instead of sitting on a lock waiting for a decision it can’t make. Trace those three files together; the protocol only makes sense as a set.
• Memory, including what does not go in it. Three active entries and one archived. Every entry passes the admission test — nothing derivable from reading the repo. One is pinned (the latency budget: its own success makes it look unused, which is why sweeps must not archive it), and the archived one shows the healthy end state — the lesson became a test, so the entry retired.
• Active context caught mid-item: what a session would need to resume BL-0002 without re-deriving it, and an explicit note on what deliberately isn’t there.

Things NOT included (intentionally)

• Source code. This is docs-only; a real adopter would have src/, tests/, etc. alongside. The paths in the Code Maps and memory entries refer to files that don’t exist here.
• CHANGELOG / README of the project itself. A real adopter would have these; we showed only the methodology artifacts.
• Project instruction file (CLAUDE.md / AGENTS.md). A real adopter starts from templates/; reproducing one here would just be the template with tinker substituted in.
• FEEDBACK.md. Deliberate: tinker is pre-alpha with no users, and the triage flow only becomes load-bearing at alpha and beyond. An empty inbox would demonstrate the file, not the practice.
• Full backlog. 5 items in one epic; a real project would have 30+ items across multiple epics.

Methodology version

Built against methodology v1.28.0.

This site is open source. Improve this page.