docs: add org-wide CONTRIBUTING, SECURITY and CODE_OF_CONDUCT

[jacderida]

Draft — for review before anything is merged.

First step on V2-878. Adds the three default community health files to WithAutonomi/.github, where GitHub inherits them across every repository in the organisation that does not define its own.

What reviewers should push back on

1. The contribution standard is test evidence, not familiarity with the code

An earlier draft of this said "if you cannot explain why a line is there and why it is correct, it should not be in your pull request". It was withdrawn, because we could not honestly hold anyone to it. Our own code is agent-generated. Demanding a line-level familiarity from an outside contributor that we do not always hold ourselves would not survive the first person who read our commit history.

What stands instead:

• You are responsible for what you submit. If it merges and it breaks, we come back to you.
• Evidence is the bar, not authorship. A PR must show what was run and on what platform, what the tests actually cover, a regression test that fails before and passes after for a bug fix, and how anything the suite cannot reach was verified.
• "It compiles and CI is green" is explicitly not evidence for a behavioural change. CI proves nothing already-tested broke; it says nothing about whether the new thing works.

This also fits the failure mode better. Generated code fails by being plausible, well-formed and subtly wrong — line-reading catches that poorly, tests catch it well. And it gives reviewers something checkable instead of a claim about a contributor's state of mind.

The amount of evidence is tied to the existing risk-tier system, rather than left as a vague exhortation. CONTRIBUTING.md now reproduces the boundary question from the PR template — does this change node behaviour, the wire protocol, the stored-data format, payments and economics, or the upgrade mechanism? — and what each tier expects:

Tier	Evidence expected
T0	Repository CI
T1	CI, plus a production compatibility smoke test
T2	A dev testnet run, plus an ADR
T3	Everything T2 requires, plus adversarial testing

It names the six repos that carry the template (ant-node, ant-client, ant-protocol, evmlib, saorsa-core, saorsa-transport), and tells contributors to repos without it to scale evidence to risk on the same basis. It also says to propose the tier yourself and that a maintainer will tell you if you pitched it too low.

Is this the right bar? It is deliberately demanding on testing and deliberately silent on authorship. Worth a second opinion — particularly on whether the tier descriptions read correctly to someone who works on the network repos daily.

2. The conduct reporting contact is a stopgap

conduct@autonomi.com does not exist, so reports are directed to the organisation owners instead, with a note to raise it with a different owner if the report concerns one.

This works today but it is not good. A dedicated, monitored alias is the right answer, and someone should decide whether to create one before this merges.

3. Linear is private, so outside contributors cannot link to it

CONTRIBUTING.md directs contributors to GitHub issues, states that Linear is internal and they are not expected to link it, and says that if a repository's linear-link check fails on an outside contribution, that is ours to fix, not theirs.

There is a live process gap behind this: on the six crate repos running check_pr.py, an external contributor cannot pass a required check without a maintainer intervening. Worth deciding whether that check should exempt contributors without write access. Not fixed here.

Coverage

A repository-level file always wins over the inherited default, so these apply to 22 of 26 public repos. Not covered:

Repo	Has its own
ant-merkle, autonomi-app-docs, autonomi-developer-docs, indelible	CONTRIBUTING.md
ant-sdk, saorsa-transport	SECURITY.md

Left alone deliberately. ant-sdk's security policy is a detailed repo-specific threat model and is better than a generic one — that should stay permanently. ant-merkle's CONTRIBUTING.md is likely inherited from the upstream rs_merkle project it was forked from and probably should be deleted so it inherits.

LICENSE is not inheritable — GitHub requires licence files in each repository so they are present when a project is cloned or packaged. That is why V2-849 needed 17 separate PRs rather than one file here.

AGENTS.md is also not inheritable; agents read the repository they are working in. Per-repo work.

Other decisions

• SECURITY.md requires reports to be reproduced by a human. Reproducing a vulnerability is a much lower bar than explaining an implementation, so this survives the objection above. It is the specific failure that closed curl's six-year bug bounty — roughly 20% of submissions were unverified model output and the confirmation rate collapsed from 15% to under 5%.
• No standalone AI_POLICY.md. Projects that have one wrote it after volume forced their hand; our inbound volume is near zero. The rules live inside CONTRIBUTING.md and can be split out if that changes.
• No link to the community forum. It is community-run rather than ours and currently offline. Discord is used instead. Note indelible's own CONTRIBUTING.md still points there.

Follow-up, not in this PR

• Create a dedicated conduct address, or confirm the owners route is acceptable
• Decide whether linear-link should exempt outside contributors
• Consolidate AGENTS.md — 8 repos have CLAUDE.md, only 3 have AGENTS.md, and AGENTS.md is the cross-tool standard read natively by Claude Code, Codex, Cursor, Copilot and Gemini CLI
• Decide whether the four repos with their own CONTRIBUTING.md should keep it

Refs V2-878