v5.4.0 — audit-driven overhaul: deploy sync, model tier v2, native orchestration, hygiene - #80
Merged
Conversation
…ive orchestration, hygiene Full-repo audit (29-agent workflow, 13 confirmed high findings) fixes: Critical fixes (A): - SKILL.md deploy: checksum sync (cmp -s) replaces copy-if-missing — rule/agent updates now reach ~/.claude and ~/.codex; hooks.json install-if-missing only, never clobbers local edits; codex/AGENTS.md now deployed at all - Phase 0: state init/handlers merge context instead of clobbering run_id (telemetry no longer silently dies); greenfield writes mode "greenfield"; codex audit prompt resolved via absolute skill path + TIMEOUT_CMD guard - Phase 2: parallel waves = N implementers + orchestrator-driven review/PAR/ship per sprint (subagents cannot dispatch subagents); STAGE_INDEXES order fixed - Phase 3: sprint-queue.json fossil removed; baseline via completion-data.json baseline_test_cmd with charter fallback - uuidgen lowercased everywhere (macOS uppercase broke event-schema contract) - Testcontainers canon: TESTCONTAINERS_RYUK_DISABLED gated on CI; cleanup by org.testcontainers=true label (name-regex matched any default-named container); helper script is the only docker command in orchestrator budget Model policy (B): - deep reviewers + deep-analyst -> fable; tier tables reconciled across enforcement/frontmatter/llms.txt; Codex deep tier xhigh; claude-opus-4-7 -> claude-fable-5 (28 occurrences); haiku opt-in for mechanical Phase 0 checks Native orchestration (C): - Reviewers emit fenced-JSON verdicts; .par-evidence.json assembled mechanically - Re-review via SendMessage to the flagging reviewer (context intact) - Monitor-based CI wait on Claude runtime; fallback chain codex -> /code-review skill -> split-focus; /code-review ultra suggest-only - Heartbeat cadence: boundaries/transitions/post-compaction on Claude, every-turn on Codex; Agent(isolation: worktree) documented as opt-in Hygiene + defense (D): - .github/workflows/ci.yml: shellcheck, DAG verify, JSON validation, forbidden-token gate; anti-injection guardrails in all 12 agent definitions - pr.fail event type (schema now 21 types); sprint-queue.json deleted; project-health-report.md marked superseded; docs refreshed to v5.4.0 Verification: verify-phase2-dag.sh 33/33 PASS; bash -n all scripts; jq all JSON; dual review (technical + coherence) -> 12 fixes -> scoped re-review APPROVE. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…fore autonomous run User picks the deep-tier model profile in Phase 1 (recommendation-led, frontier default), can change it at plan approval; persisted to .superflow-state.json context.model_profile + Autonomy Charter; dispatch rules and Codex secondary pins are profile-aware. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Implements all fixes from the full-repo audit (29-agent workflow: 7 subsystem auditors, 4 gap analysts, adversarial verification — 13 confirmed high findings).
A. Critical fixes
cmp -s) redeploy in SKILL.md startup replaces copy-if-missing — enforcement-rule and agent-definition updates now actually reach~/.claude/rules/,~/.claude/agents/,~/.codex/;hooks.jsoninstalled only if missing (local edits never clobbered);codex/AGENTS.mdnow deployed at all.context.run_id(telemetry no longer silently dies); greenfield writesmode: "greenfield"(crash recovery no longer re-enables Branch A over fresh docs); codex audit prompt via absolute skill path +TIMEOUT_CMDguard.sprint-queue.jsonfossil dependency removed; baseline test viacompletion-data.json: baseline_test_cmdwith charter fallback.uuidgenbroke the event-schema run_id contract).TESTCONTAINERS_RYUK_DISABLEDgated on CI; label-based cleanup (org.testcontainers=true) — the old name-regex matched any default-named docker container, incl. dev databases.B. Model policy
Deep reviewers + deep-analyst →
fable; tier tables reconciled across enforcement / agent frontmatter / llms.txt (they disagreed in 3 ways); Codex deep tierxhigh;claude-opus-4-7→claude-fable-5; haiku opt-in for mechanical Phase 0 checks.C. Native orchestration
Fenced-JSON reviewer verdicts →
.par-evidence.jsonassembled mechanically; re-review viaSendMessageto the flagging reviewer; Monitor-based CI wait (Claude runtime); fallback chain codex →/code-reviewskill → split-focus; heartbeat cadence split (boundaries on Claude, every-turn on Codex);Agent(isolation: worktree)documented as opt-in.D. Hygiene + defense
Repo CI (
.github/workflows/ci.yml: shellcheck, DAG verify, JSON validation, forbidden-token gate); anti-injection guardrails in all 12 agent definitions;pr.failevent type (21 total); fossils deleted/superseded; CLAUDE.md / llms.txt / CHANGELOG refreshed to v5.4.0.Test plan
tools/verify-phase2-dag.sh— 33/33 PASSbash -non all shell scripts;jq emptyon all JSONclaude-opus-4-7, bareRYUK_DISABLED, copy-if-missing guards)🤖 Generated with Claude Code