chore: ignore gitnexus-generated agent files

[MichaelTaylor3d]

chore: ignore gitnexus-generated agent files

gitnexus analyze writes agent tooling into the repository it indexes, and sibling repos committed it; this repo has not, and this ignore keeps it that way. CLAUDE.md is deliberately still tracked because it is hand-authored, so analyze must be run with --skip-agents-md here.

Refs #3177

[MichaelTaylor3d]

loop-reviewer verdict: PASS

Head SHA reviewed: f03233d

Verified independently (fresh context, own git objects / GitHub API reads):

1. /CLAUDE.md is ABSENT from the ignore block. Confirmed by fetching CLAUDE.md at this repo's main — it is the hand-authored 44-line "docs.dig.net — repo operating notes" file, zero gitnexus markers, unaffected by this PR. PASS.
2. Diff touches exactly one file, .gitignore. gh api .../compare/main...f03233d5 reports files: [".gitignore"], status: ahead, no deletions beyond the .dig line being re-emitted with a trailing newline. PASS.
3. No line names DIG-Network/dig_ecosystem or #3177 — grepped the diff, zero matches. PASS.
4. Trailing-newline change is benign and isolated: byte-level diff of .gitignore at base vs head shows .dig (no trailing LF) becomes .dig\n, followed by a clean 9-line LF-terminated append (comment block + 4 ignore paths). No CRLF anywhere, no other line's terminator touched. PASS.
5. Zero intersection: fetched origin/main's full tree via GitHub API recursive listing and grepped for AGENTS.md, .claude/skills/gitnexus/, .claude/skills/generated/, .gitnexus/ — no matches in either direction. The new ignore patterns shadow nothing tracked. PASS.
6. Fast-forward confirmed: GitHub compare API reports ahead_by: 1, behind_by: 0 relative to main — clean FF descendant, no rewrite. PASS.
7. Commit message and PR title are both chore: ignore gitnexus-generated agent files — conventional-commit shaped (chore: type, imperative description), passes commitlint's PR-title lint. PASS.

All 7 checks pass. No findings, no threads to open.

Note on approval: --approve was not attempted — this PR's sole commit is authored by the same identity this review runs as (michael@michaeltaylor.dev / MichaelTaylor3d), and GitHub structurally rejects self-approval ("Can not approve your own pull request"). Posting via --comment instead per brief; this PASS should be treated as the equivalent gate signal for merge purposes, decided by the orchestrator/human with merge rights.

[MichaelTaylor3d]

loop-security pre-merge audit — CHANGES-REQUIRED

Head audited: f03233d505223b84ca3f9666d1b8e84150c4bac9 (base main @ 59fea7edb3fd7a6c0d57ab72ca3b750822bd1ed0, single commit, fast-forward, mergeable: MERGEABLE)

Verdict sentence: The added .gitignore text itself discloses nothing new (no hostnames, repo names or credentials in the patch), but this PR's commit message and PR body both still contain Refs #3177 — the same internal ticket number that the earlier, already-scrubbed revision of this exact fix leaked as DIG-Network/dig_ecosystem#3177 across 16 public files. It is not a secret, and it does not name the private repo, but it is a live recurrence of the disclosure pattern this fix exists to eliminate, in the two places (commit message, PR body) that a squash-merge will carry verbatim into the public repo's permanent history.

Seven checks

1. New disclosure in the added text — FAIL. grep -c 'dig_ecosystem' .gitignore = 0 and no #3177 in the file body (confirmed via GitHub contents API at this head). But the commit message (git log/API: author: Michael Taylor <michael@michaeltaylor.dev>, message ends Refs #3177) and the PR body (identical text, ends Refs #3177) are NOT clean — this is exactly the check the brief called out as having failed once already.
2. Nothing deleted — PASS. Exactly one file changed (.gitignore), diff shows +10/-1; the single deletion is the pre-existing .dig\ No newline at end of file line, replaced by .dig + newline + the new block. No path removed.
3. /CLAUDE.md absent from the ignore block — PASS. Confirmed by reading the full patch text via the GitHub diff API; only /AGENTS.md, /.claude/skills/gitnexus/, /.claude/skills/generated/, /.gitnexus/ are added. CLAUDE.md remains tracked in git ls-tree -r main (confirmed present).
4. No security-relevant path caught by the new patterns — PASS. Intersected AGENTS.md, .claude/skills/gitnexus, .claude/skills/generated, .gitnexus against git ls-tree -r --name-only main (recursive tree via API): zero matches for all four; only CLAUDE.md (correctly excluded from the ignore block) is present.
5. No git-identity manipulation — PASS. author: Michael Taylor <michael@michaeltaylor.dev> | committer: Michael Taylor <michael@michaeltaylor.dev> — the inherited machine identity, both fields match.
6. No rewritten history — PASS. Single commit, one parent (59fea7e...), matches baseRefOid; mergeStateStatus: MERGEABLE, a clean fast-forward descendant of main.
7. Line-terminator integrity — PASS. Patch hunk is a genuine 9-line append (+10/-1 net) touching only the tail of the file to fix the missing trailing newline on .dig; the rest of the file (Dependencies/Production/Generated/Test-artifacts/Misc blocks above) is untouched in the diff — not a whole-file CRLF rewrite.

Finding

• f03233d5 commit message + PR body, line "Refs #3177" — PRIVATE-BUT-HARMLESS, but a live recurrence of the exact defect this PR fixes. No exploit beyond ticket-number correlation (an outside reader who also saw the now-removed DIG-Network/dig_ecosystem#3177 string in a sibling repo, or finds it in this repo's future git blame/PR history, can confirm this hygiene PR belongs to the same private internal batch). Not a secret, no host/URL/credential. Recommend: edit the PR title/body to drop Refs #3177 before merge (GitHub defaults the squash-commit message to the PR title+body, so this text will otherwise land permanently in main's history) and merge with a scrubbed message instead of the raw commit message.

What I could not check

• Could not check the local working tree / CRLF byte-for-byte (audited via GitHub's diff and contents APIs only, read-only, no local clone/worktree touched — nothing to restore).
• Did not attempt gh pr review --approve (same identity as PR author — expected to fail per brief); posting this as a PR comment instead.

KG: private-but-harmless-recurrence — a scrubbed leak's ticket number survives in the commit message/PR body even after the file body is fixed; squash-merge defaults propagate PR title+body verbatim, so the merge commit needs its own scrub, not just the diff.