North-Star Outcomes

Code-mint is most useful when the user can see a small number of concrete capabilities getting unlocked over time. These outcomes are the public promise of the onboarding flow.

The Outcomes

Outcome	What It Means	Proof Of Completion	Primary Skills
Validate Current State	The agent can assess the repo as it exists today instead of guessing.	All applicable baseline audit outputs are recorded, and a concise summary captures what is working, blocked, risky, and next to prove.	meta-onboarding, applicable auditor skills
Navigate	The agent can explain the repo structure, identify the right module for a sample task, and justify where work should happen.	Durable in-repo guidance plus a grounded repo explanation or sample task walkthrough that shows the map is actually usable.	legibility-auditor, legibility-enhancer
Self-Test	The agent can run the smallest relevant automated check for a real module or behavior and trust the result.	An exact targeted test command or test target, the scope it covers, and a trustworthy pass/fail signal are captured in the checklist.	autonomy-testing-auditor, autonomy-testing-creator
Smoke Path	The agent can execute one safe, non-destructive runtime confidence check.	A documented smoke path with prerequisites, exact steps, stop conditions, and a concrete success signal such as a response, log line, or health check.	autonomy-env-auditor, autonomy-env-creator, autonomy-staging-auditor, autonomy-staging-creator
Bug Reproduction	The agent can turn a real reported issue into a reproducible failing case before proposing a fix.	A failing test, script, or deterministic reproduction recipe tied to the reported issue and runnable by another person or agent.	interface-sre-agent, autonomy-testing-*
SRE Investigation	The agent can inspect logs, metrics, traces, CI, or infra context and produce a ranked hypothesis.	A short investigation log captures the evidence sources used, observed signals, ranked hypothesis, and next actions, or records why the outcome is N/A.	autonomy-sre-auditor, interface-sre-agent

Why These Outcomes

These outcomes map to three pillars:

1. Legibility — durable guidance so an agent can navigate intentionally.
2. Autonomy — trustworthy tests, safe runtime checks, and where applicable operational CLI access (see docs/framework.md for what “autonomy” does and does not promise).
3. Human-AI Interface — the bidirectional channel between humans and the system, mediated by the agent. Inbound: interface-ticket-writer refines vague requests into executable work. Outbound: interface-sre-agent interprets operational signals and surfaces them back to humans.

Interface skills are activated after the core onboarding outcomes are in place. Once the repo is legible and verifiable, better ticket refinement and signal interpretation compound the rest of the system.

Evidence Model

Every outcome should be tracked with the same evidence fields:

• Status: Not Started, Blocked, In Progress, Proven, or N/A
• Evidence: the artifact that proves the outcome, such as a report path, AGENTS.md, test name, screenshot, prompt transcript, or investigation note
• Exact Check: the command, prompt, or scenario that produced the evidence
• Last Validated: when the evidence was last refreshed
• Notes: what is still manual, flaky, blocked, approval-gated, or intentionally marked N/A

Use docs/onboarding-checklist.md as the canonical template for that evidence. .agents/code-mint-status.json mirrors outcome statuses as a machine-readable index for cross-repo scanning. For an optional skill-by-skill matrix aligned to outcomes, see docs/skills-status.md.

What "Done" Looks Like

An onboarding effort is working when the user can point to a short checklist and say:

• the current state has been baselined with recorded audit evidence
• the repo map can answer where work should happen for a sample task
• the agent can run the smallest relevant automated check and trust the result
• the agent can perform one trusted smoke check with a concrete success signal
• a reported issue can be turned into a deterministic failing repro
• operational evidence can be gathered into a ranked investigation hypothesis

That is the story the rest of the repository should reinforce.