Loading…
Loading…
dod · an Agent Skill by SkillEra for Claude Code, Codex, Cursor and any agent that reads SKILL.md
Plan a feature layer by layer, get the plan independently reviewed, freeze it, track the build against it, and close with a number: how much of the design the plan foresaw.
Report · 2026-09-15
Most features are planned as their base functionality. The rest — the empty state, the second user, the abusive input, the dependency that is down, the rollback — is discovered during the build, in cycles of revision.
dod front-loads that discovery. It walks a fixed set of consideration layers, refuses to call a plan ready until every applicable layer has an evidenced answer and an independent reviewer agrees, then tracks the build against the plan and reports how much of the design the plan foresaw.
By default it runs only when you ask: invoke /dod, name the skill, or ask for a definition of done for a specific feature. An ordinary “make a plan” does not trigger it unless the project has opted into automatic triggering, and it never interrupts to ask whether it should.
Described from the README at dod-v0.2.0 (opens in a new tab).
1
S, M, L or Epic. Smaller than S — a typo, a one-line fix — and the skill says so and stops.
2
It reads the modules the feature touches, the tests, the schema, auth and routing, and records the commit it read. It never asks what the code can answer.
3
Fifteen layers, forty-nine probes. Each probe is answered from the brief and the recon with a pointer into the plan, or it is a Gap.
4
One batch, at most about eight, gaps only — each with why it matters and a recommendation you can accept in one word.
5
The Definition of Done comes first: verifiable statements D1…Dn, each with an evidence type — test, cmd, file or manual — then a build plan an agent that never saw the conversation can execute.
6
A reviewer gets a score-redacted copy and the rubric, and returns findings with READY or REVISE and its own coverage line. After three rounds the plan comes to you, with both coverage lines side by side, to decide.
7
Ready needs every applicable layer and probe, no gating probe open, and READY on the current revision. The Definition of Done is frozen into a Baseline that is never edited again.
8
Follow the build plan; check an item only with evidence; anything unforeseen is a typed amendment before it is built.
9
Every item verified, then the report: the prediction rate — baseline items over baseline items plus the discovered and corrected amendments, the changes the plan should have caught.
Every applicable layer needs an evidenced answer before a plan can be approved. Seven probes are gating: a plan with one of them open is never ready.
layer holds a gating probe
In Markdown, under version control. There is no database and no service. The author never grades alone: the reviewer is the Codex CLI run read-only, a fresh-context subagent, or you.
Then run /dod setup in a project. It creates the plan store and shows you the pointer block before writing it into your agent's instructions file — nothing is written silently.
Or copy the skill folder to ~/.claude/skills/dod/, ~/.codex/skills/dod/ or ~/.cursor/skills/dod/. The helper script needs Node 20 or later.
Claude Code, Codex, Cursor and others, through the skills installer.
npx skills add KDavidP1987/dod-skillAdd the marketplace once, then install the plugin.
/plugin marketplace add KDavidP1987/dod-skill
/plugin install dod@dod-skillThe walkthrough follows one plan through five review rounds to its close. The skill is MIT licensed; issues and pull requests are welcome on GitHub.