Codex Bridge

Reads .codex/ at invocation time and composes a briefing the way ./bin/hushh codex route-task does: workflow → owner_skill + default_spoke → union of required_reads, required_commands, handoff_chain, verification_bundle, risk_tags.

Response rules

The repository routing contracts are the source of truth. The enforceable rules live in .codex/skills/comms-community/references/reply-rules.md and are injected into the routed briefing when the turn is a community reply — the route lands on comms-community / community-response, or the prompt names the channel outright (see scripts/route.py). They are deliberately not injected into engineering questions: they are a 180-line Discord tone contract, and a question mark in a bug report is not a request for Discord voice. Follow that file literally when it does load. If it conflicts with anything written here, it wins.

Host invocation guidance:

1. When you cannot reach .codex/ to fetch the routing contracts, report that routing is unavailable and inherit the communication rules in root AGENTS.md. Do not substitute a channel-specific style or fabricate a routed lane.
2. Pull the signature values (<skill-id>, <workflow-id>) from the routed briefing header (# Routed workflow: ... / # Routed skill: ...).
3. If the briefing header shows a disambiguation table or a catalog, do not fabricate a signature, invoke the bridge again with the chosen name first.

Anything stylistic beyond this belongs in reply-rules.md, not here. Edit the codex file and the bridge will pick it up on the next invocation.

Routed briefing

!python3 ${CLAUDE_SKILL_DIR}/scripts/route.py $ARGUMENTS

How to execute what's above

A single skill or workflow briefing

Treat it as the instruction set:

1. Scope check. Honor the skill's "Do Use" / "Do Not Use", primary_scope, owned_paths. If the task falls outside, invoke /codex-bridge <handoff-target> instead of improvising.
2. Read first. Open every .md under "Read First (composed)" before touching code. The composed list already includes the workflow's, the owner's, and the spoke's required_reads.
3. Follow the Workflow / Playbook section verbatim. That's what codex has already decided works.
4. Run the composed Required Checks before declaring done. The Bash(python3 *) grant covers Python-based checks; other bash tools will prompt once.
5. Hand off on drift. If work expands, stop and /codex-bridge <next> — usually the next entry in handoff_chain.

A compact pointer (Routed lane (compact))

The match was plausible but not confident, so the bridge sent the pointer instead
of the full playbook: lane name, owner/spoke, read-first paths, required checks,
and the runner-up candidates. Treat it as a suggestion. If it is the right lane,
run /codex-bridge <name> to load the full briefing; if it is not, ignore it and
proceed with the user's request as stated.

An ambiguity / disambiguation table

Multiple skills scored close. Prefer the spoke over the owner when both match; prefer a workflow over a bare skill when a workflow is listed (workflows compose across owner + spoke).

A catalog

No strong match. Pick by description, re-invoke /codex-bridge <name>.

An agent briefing

The matched entry is a repo-scoped custom agent (under .codex/agents/), not a skill or workflow. Treat it as a delegation lane, not instructions to execute directly:

1. Decide whether the current turn actually benefits from delegation. Most single-lane requests do not.
2. If delegation fits, confirm with the user before invoking the corresponding /agent <name>. Subagent use is explicit only per .codex/skills/agent-orchestration-governance/references/delegation-contract.md.
3. When the briefing appears as a Suggested delegation lanes footer on a skill or workflow briefing, treat it as context only. Execute the primary briefing normally.

A --check report

Surface findings to the user. Don't auto-fix — --check is a health report, not a migration.

Invocation patterns

/codex-bridge                                 # catalog of owners, spokes, workflows, agents
/codex-bridge backend-api-contracts           # compose skill briefing
/codex-bridge security-consent-audit          # compose workflow briefing (owner+spoke unioned)
/codex-bridge governor                        # compose agent briefing (delegation lane)
/codex-bridge "how does Kai maintain session continuity"  # free-text → token scoring
/codex-bridge --check                         # structural lint of .codex tree

Design

• Source of truth: .codex/skills/*, .codex/workflows/*, and .codex/agents/*. Edit there; the bridge re-reads every invocation.
• Composition mirrors codex. Routing emits the same union of fields that repo_scan.py::build_route_task produces, so Claude makes the same scope decisions codex would.
• Agents are advisory lanes, not winners. Exact agent name routes to an agent briefing. Free-text never elevates an agent above a matching skill or workflow. On explicit skill or workflow invocations, matching agents appear as a compact Suggested delegation lanes footer (suppressed on Q&A turns and on close-scoring ties).
• Progressive disclosure. Only the routed briefing enters context, not the full corpus. The --hook path adds a second axis: silence below the gate, a compact pointer at medium confidence or when a rival lane scores within 2 points, and the full composed briefing only when the match is both strong and unambiguous.
• Rarity-weighted matching. Token overlap is weighted by how rare the token is in the corpus, so shared words (audit, review, contract) stop deciding the route and a path the user actually typed outranks any amount of prose overlap. Fixtures in fixtures/routing_cases.json pin the expected lane for representative prompts and cap total injected size; scripts/test_route_hook.py enforces both in CI.
• Path-aware auto-trigger. paths: .codex/**, consent-protocol/**, hushh-webapp/**, docs/** covers the repo surfaces where a specialist applies.