Skip to content

feat: harness-name prevention gate + name-banlist intent (itd-74) - #1

Merged
REPPL merged 1 commit into
mainfrom
feat/harness-name-prevention-gate
Jul 7, 2026
Merged

feat: harness-name prevention gate + name-banlist intent (itd-74)#1
REPPL merged 1 commit into
mainfrom
feat/harness-name-prevention-gate

Conversation

@REPPL

@REPPL REPPL commented Jul 7, 2026

Copy link
Copy Markdown
Owner

The deterministic guard so banned names can't creep back into published content — the generalised follow-up to the recent host-agnostic cleanup.

Two layers, split by sensitivity

Public (CI-enforced) — a harness/* banned-token family in .abcd/docs-lint.json: specific agent-harness names are blockers in user-facing content. README.md is now scanned alongside docs/; a per-line <!-- docs-lint: allow --> comment is the escape hatch. The .abcd/ design record is deliberately not scanned (it records real integration decisions).

Private (local-only) — a committed .githooks/pre-commit (with no banned name in it) reads an untracked, gitignored banlist under .abcd/.work.local/ and refuses to stage matching content. This is how a name that must never appear in any public artefact is enforced without its literal string ever entering tracked content or CI config — the case a public linter rule structurally cannot cover. No-op on machines without the banlist (fresh clones, CI).

Tests / verification

  • New TestDocsLintHarnessNameGate loads the real config and asserts the family flags a named harness at the right line and honours the allow-context — so deleting the family (or downgrading its severity) fails CI.
  • Guard verified both ways (blocks a banned name, passes clean content); it ran and passed on this very commit.
  • build / vet / test / record-lint / docs lint all green; gofmt clean.

itd-74

Draft intent capturing the generalisation: abcd manages both banlists as first-class config and ahoy scaffolds them for any managed repo. This repo is the working prototype.

Assisted-by: Claude

Two-layer guard against reintroducing banned names:
- Public (CI-enforced): a harness/* banned-token family in the docs-currency lint
  (.abcd/docs-lint.json) — specific agent-harness names are blockers in user-facing
  content, with README now scanned alongside docs/ and a per-line
  <!-- docs-lint: allow --> escape hatch. A regression test loads the real config
  and asserts the family bites and respects the escape.
- Private (local-only): a committed .githooks/pre-commit guard reads an untracked,
  gitignored banlist (.abcd/.work.local/) and refuses to stage matching content, so
  a name that must never appear in any public artefact is enforced without its
  literal string ever entering tracked content or CI config.

itd-74 captures the generalisation (abcd manages both banlists; ahoy scaffolds them
for any managed repo). This repo is the dogfood.
@REPPL
REPPL merged commit 5290e82 into main Jul 7, 2026
10 checks passed
@REPPL
REPPL deleted the feat/harness-name-prevention-gate branch July 7, 2026 11:00
REPPL added a commit that referenced this pull request Jul 11, 2026
feat: harness-name prevention gate + name-banlist intent (itd-74)
REPPL added a commit that referenced this pull request Jul 27, 2026
The registry is the itd-103 core: entries drawn only from this repo's own
documented never-list, each carrying a command-position pattern over shell
tokens, a blocker/warn tier mirroring the record-lint severity family, a safe
successor, and a plain-language why — because the refusal itself is the lesson.

Check returns allow / warn / block as values only; no stdout, no exit status.
Fail-open-loud on a broken guard is the hook shim's job, so core only names the
failure (ErrUnparsableCommand) and lets the caller decide.

Matching is command-position only, so a hazard named inside a quoted argument
never fires, while quoting is not a bypass and wrapper or environment prefixes
do not hide the command. The admission gate test holds every bundled entry to
its proof: each known-bad fixture fires at the declared tier, every known-good
fixture fires nothing (100% true-negative floor), and known-good is at least
40% of the corpus. Known-good fixture #1 is this repo's own incident capture,
the command a raw-regex guard would have refused.

Assisted-by: Claude:claude-opus-5[1m]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant