v1.8

Default model bump + versioning/release hygiene.

Changes

• Default model → Opus 4.8. The built-in default for the reasoning stages (Investigator + Critic) is now us.anthropic.claude-opus-4-8. The Reporter and issue-answer paths stay on Haiku 4.5 (Opus rejects structured output over Bedrock). No caller change required — adopters already overriding BEDROCK_MODEL_ID are unaffected. Every default now resolves from a single source of truth (config._DEFAULT_MODEL) so provenance, pricing, and the doctor can't drift.
• v0 frozen; v1.x is the release track. The old moving v0 tag is frozen at its final pre-v1.x revision and no longer advances. Pin a numbered v1.x tag (or its commit SHA for the strongest supply-chain guarantee) — see the README 'Supply-chain pinning' section. All install docs, examples, the reusable-workflow default, and the CFN Launch Stack URL now point at a numbered release.
• Docs: added the BEDROCK_ISSUE_MODEL_ID env row; honest tag-vs-SHA guidance; dropped the refs/tags/v1.* OIDC-trust wildcard suggestion (it would auto-trust future tags).

Notes

• 541 tests pass. Reviewed across 10 dimensions (correctness, security, contract-drift, comment-rot) with the findings applied.
• Engine ref: 5f3386a83068d809f901c244462c55d17df166d2
• Pinning: uses: sudsali/shadow/.github/workflows/shadow-review.yml@v1.8 (or the SHA above).

v1.7

Adds two configurable engine features. No behavior change unless opted into via .shadow.yml.

Changes

• Configurable allowed_labels — the set of labels issue-triage may apply is now tunable via bot.allowed_labels in .shadow.yml (or the BOT_ALLOWED_LABELS env var). Defaults to bug/enhancement/question/documentation/help-wanted; model-proposed labels outside the set are dropped.
• Brand-aware rate-limit label — the rate-limit escalation label is now derived from bot.name (<bot.name>:rate-limited) instead of a hardcoded shadow:rate-limited, so it's on-brand on repos that rename the bot.

Engine ref: ed31ed7e30ef81be82d7b8b756b626f876036d73

Pinning: uses: sudsali/shadow/.github/workflows/shadow-review.yml@v1.7 (or the SHA above).

v1.6

Decouples the issue-answer model from the PR Reporter model.

Changes

• Thread a new BEDROCK_ISSUE_MODEL_ID through the reusable workflow so adopters can answer issues/followups on a different model than the PR-review Reporter, without moving the Reporter off Haiku. Defaults to the Reporter model (Haiku 4.5) when unset — zero behavior change until opted into.

Engine ref: 54ec94e0ca8c90d9b58ff95a2a06b175a115784e

Pinning: uses: sudsali/shadow/.github/workflows/shadow-review.yml@v1.6 (or the SHA above).