Agentic Coding

Know where you are. Make one move. Read what changed.

Agentic Coding gives an AI repository and tool-state proprioception: where it is, what changed, what evidence is still needed, and where it must stop. It keeps live coding work inside a small evidence loop instead of confusing code volume, tool success, or repeated attempts with progress.

Open the project site →

This repository contains the curated contest skill shipped with Nova, copied from the public Nova + MIND OpenAI Build Week release into a fresh standalone history. Private development history is excluded.

• Contest edition: 1.0.0
• Skill: SKILL.md
• License: MIT
• Contest source: Agentic Coding in Nova

This is a clean standalone source link. Independent plugin installation is not claimed by the contest evidence.

The evidence loop

Agentic Coding enters through the repository that exists now and lets each move earn the next one with a new observable:

map -> localize -> hypothesize -> make one bounded probe or change
    -> interrogate an oracle -> interpret -> checkpoint, recover, or hand off

The operating posture is deliberately narrow:

1. Map the repository boundary, current state, local instructions, and behavior surface.
2. Localize the target beside its nearest tests, callers, contracts, configuration, and error path.
3. Hypothesize one causal mechanism and name the observation that would falsify it.
4. Probe with the smallest reversible change or experiment that can discriminate that hypothesis.
5. Interrogate the narrowest available test, build, lint, typecheck, or target readback.
6. Interpret what the result supports—and what its scope does not establish.
7. Checkpoint, recover, or hand off with an exact re-entry condition.

A useful invocation

Use $agentic-coding to diagnose this failing repository behavior.
First map the live state and local instructions, then identify one causal
hypothesis with a falsifier. Make only the smallest authorized probe or change,
run the narrowest discriminating oracle, and report what the result supports,
the unresolved risk, and the exact re-entry condition.

The skill is designed for live repository work inside a host that already provides the relevant files, tools, and action boundary.

The work packet

When an episode crosses turns, tools, mutations, failure recovery, or handoff, preserve a compact resumption surface containing:

Field	Record
Objective	Intended behavior and the evidence that would advance acceptance.
Authority	Explicit permission, included scope, and excluded scope.
Repository state	Root, branch or working state, local instructions, and known mutations.
Decisive facts	Tests, callers, interfaces, configuration, and error-path observations that constrain the explanation.
Live hypothesis	One causal story paired with a falsifier.
Next probe	The smallest reversible information-producing move and its exact re-entry condition.

A work packet is not a transcript. Its value is competent resumption without rediscovering the repository or laundering stale assumptions into current truth.

Interrogate the right oracle

Choose the narrowest check that can actually disagree with the live hypothesis:

Oracle	What it can establish
Targeted test	A meaningful before/after signal for the defect or requested behavior.
Build	Integration, generated output, bundling, and configuration surfaces.
Lint	Static policy and convention compliance—not runtime behavior.
Typecheck	Contract compatibility and data-shape assumptions—not successful execution.
Target readback	The actual state written or changed, rather than a tool's success message.

Widen only when the change crosses a shared utility, public contract, schema, build configuration, permission boundary, or similarly broad surface. A passing command stays scoped to the behavior it exercised.

Recover by changing the premise

Preserve the failure signature and the last-known-good state. Then classify the episode before another attempt:

• environmental fault;
• pre-existing failure;
• false hypothesis;
• patch regression;
• insufficient oracle;
• unknown mutation state;
• another explicitly evidence-bearing condition.

Change the localization, premise, probe, or oracle before retrying. Repeating the same move after the same observation is not recovery.

Return a truthful technical record

Close the technical pass with:

• inspected facts and touched paths;
• intended behavior and observed diff boundary;
• commands, outcomes, and the claims each outcome supports;
• unresolved risk or evidence gap;
• mission and episode identifiers, system and adapter identity where supplied, authority source, and recipient;
• the exact next re-entry condition.

At a permission, tool-availability, local-model, independent-verification, or external-consequence boundary, keep three facts separate:

1. what was authorized;
2. what was actually attempted and observed;
3. what an independent oracle accepted.

Agentic Coding is the technical repository operation inside the controlling action custody. It does not become a parallel execution authority merely because it can edit code.