claude-glm-executor-handoff
Factory-tier skill defining the input-prompt contract and GLM behavioural deltas for an in-aif Claude coordinator dispatching an executable task to a GLM-5.3 worker — a thin, cc-only adapter.
claude-glm-executor-handoff
Status: shipped-beta · Ships to: factory tier arm (GETFF_SKILLS_FACTORY, setup.d/lib.sh:63) · Fires at: Claude Code skill auto-activation when an in-aif coordinator is about to dispatch to a GLM worker ("GLM executor", "parsing a GLM worker's REPORT", …)
What it is
The cross-model dispatch edge: "Use when an in-aif Claude coordinator is about to dispatch an executable task to a GLM-5.3 worker (any agent whose frontmatter carries model: glm-5.3 or a GLM-family model). … NOT for Claude→Claude worker dispatch (use SDD directly)" (.claude/skills/claude-glm-executor-handoff/SKILL.md:3). Note the executor tier moved to glm-5.3 "since 2026-09-11" per the skill's own header — the census (pin a1337cb301) still recorded GLM-5.2; this page documents the re-probed pin (W2 delta row, §1 of the stage report). The skill is "A thin adapter for the narrow case where an in-aif Claude coordinator dispatches to a GLM worker (executor tier glm-5.3)" and owns "only the prompt side" — aif's per-agent model: frontmatter decides which model spawns (.claude/skills/claude-glm-executor-handoff/SKILL.md:13).
How it works
Sections: §0 when it fires (coordinator on Claude-family AND worker frontmatter model: GLM-family; a ZCode delivery constraint names plugin-channel agents as diagnosticOnly there) (.claude/skills/claude-glm-executor-handoff/SKILL.md:15 heading), §1 verified GLM facts with a refuted-folklore sidebar (line 31), §2 the input contract for the GLM edge (line 51), §3 "Status translation (GLM reply → orchestrator REPORT)" (.claude/skills/claude-glm-executor-handoff/SKILL.md:76), §4 recovery protocol (line 89), §5 "Honest gaps — designed-not-proven" (.claude/skills/claude-glm-executor-handoff/SKILL.md:112). Harness posture: cc-only — "requires an in-aif CC coordinator + aif runtime-bridge + GLM worker; without the bridge/worker it cannot run" (.claude/skills/claude-glm-executor-handoff/SKILL.md:6).
Satellites & companions
- USES:
superpowers:subagent-driven-development(the dispatch loop it adapts); aif's per-agentmodel:frontmatter mechanics. - ADAPTS: the orchestrator REPORT schema (owned by
agents/orchestrator-worker-discipline.md, census C1) for GLM replies; the relative-tier model posture owned bynight-mode(census B6). - ADDS: the GLM-specific input-prompt contract, verified behavioural deltas (text-only I/O,
reasoning_effortvalue-collapse, Anthropic-compat endpoint mechanics), and the status-translation table. - Census family satellites: H (runtime-bridge).
Anchors
.claude/skills/claude-glm-executor-handoff/SKILL.md:2—name: claude-glm-executor-handoff.claude/skills/claude-glm-executor-handoff/SKILL.md:3—description: Use when an in-aif Claude coordinator is about to dispatch an executable task to a GLM-5.3 worker … NOT for Claude→Claude worker dispatch (use SDD directly)..claude/skills/claude-glm-executor-handoff/SKILL.md:6—<!-- @harness-posture: cc-only — factory-depth thin adapter: requires an in-aif CC coordinator + aif runtime-bridge + GLM worker; … -->.claude/skills/claude-glm-executor-handoff/SKILL.md:13—A **thin adapter** for the narrow case where an in-aif Claude coordinator dispatches to a GLM worker (executor tier glm-5.3). … this owns only the prompt side..claude/skills/claude-glm-executor-handoff/SKILL.md:76—## §3 Status translation (GLM reply → orchestrator REPORT).claude/skills/claude-glm-executor-handoff/SKILL.md:112—## §5 Honest gaps — designed-not-provensetup.d/lib.sh:63—GETFF_SKILLS_FACTORY="dispatcher aif-doctor harvest story claude-glm-executor-handoff"
All anchors at framework pin b069c59328aa8e08671d3a19fd25efa618dfc1c5 (2026-09-11).
story
Factory-tier skill that recaps a finished session as a plain-language story told by acts — localized via AIF_HOOK_LANG and shared as one spec with the Stop-hook auto-emission branch.
tool-bootstrapping
Always-shipped skill codifying the six-rule project-aware MCP/skill proposal discipline — stack analysis, capped proposals, bulk confirmation, token-economy gate, incrementality, persistence.