Issues

fuzz: crash in skill_frontmatter
bug-ops/zeph#6805 · opened on Sep 28, 2026
research(llm): benchmark Jev (TypeSafe AI) as a fast-tier classification provider for zeph-llm
bug-ops/zeph#6801 · opened on Sep 22, 2026
1
raw self-reported LLM confidence used as uncalibrated hard threshold in correction-feedback classifier
bug-ops/zeph#6800 · opened on Sep 22, 2026
core: BudgetHint.remaining_tool_calls is static for the whole turn, not re-rendered per tool-loop iteration
bug-ops/zeph#6786 · opened on Sep 8, 2026
Claude request builder's empty-blocks guard can promote a non-user message to be first in the request
bug-ops/zeph#6784 · opened on Sep 8, 2026
hard compaction / deferred-summary hiding can orphan a tool pair with no repair outside Claude
bug-ops/zeph#6782 · opened on Sep 8, 2026
shared-state block rendering allows forged closing tag mid-line (< > / not JSON-escaped)
bug-ops/zeph#6778 · opened on Sep 8, 2026
Implement: skill execution state (SKILL.state bounded state for zeph-skills/zeph-core)
bug-ops/zeph#6766 · opened on Sep 5, 2026
research(sub-agents/orchestration): bounded structured state for long-running subagent sessions and DAG nodes (arXiv:2608.26263)
bug-ops/zeph#6750 · opened on Sep 2, 2026
1
research: zeph-agent-tools and zeph-agent-context have no dedicated top-level spec
bug-ops/zeph#6637 · opened on Jul 21, 2026
research(plugins): declared filesystem/secrets/network capability boundaries for plugin manifests (IronClaw reborn-substrate parity, ironclaw-v0.28.0)
bug-ops/zeph#6564 · opened on Jul 20, 2026
research(context): compaction has no cache-prefix-stability awareness (TokenPilot-inspired, arXiv:2606.17016)
bug-ops/zeph#6563 · opened on Jul 20, 2026
1
Track deferred architecture markers (A1-A3, D1, D2/G3, D4) from PR #3401
bug-ops/zeph#6555 · opened on Jul 20, 2026
zeph-agent-tools: re-evaluate AgentChannel dispatcher-extraction plan if revived
bug-ops/zeph#6536 · opened on Jul 20, 2026
research(tools): DeltaBox-style millisecond COW sandbox checkpoint/rollback (arXiv:2605.22781)
bug-ops/zeph#6470 · opened on Jul 19, 2026
durable TUI: map CipherError::UnknownKeyId in a future TUI reveal path
bug-ops/zeph#6459 · opened on Jul 18, 2026
durable: cooperative cancel for a live-owned execution (FR-007)
bug-ops/zeph#6440 · opened on Jul 18, 2026
parity(cli): zeph doctor is read-only advisory only — Claude Code's /doctor now diagnoses AND fixes
bug-ops/zeph#6357 · opened on Jul 17, 2026
research(context): agent-visible dashboard + reversible demand-paged eviction for context compaction (arXiv:2603.09023, arXiv:2606.30005)
bug-ops/zeph#6356 · opened on Jul 17, 2026
research(mcp): elicitation protocol redesign (SSE push → InputRequiredResult/requestState) in 2026-07-28 spec RC
bug-ops/zeph#6348 · opened on Jul 17, 2026
acp: migrate AcpPermissionGate to DurablePromise for crash-resumable IDE approvals
bug-ops/zeph#6235 · opened on Jul 13, 2026
1
orchestration: imperative mid-execution HITL interrupt tool (request_human_input) — Phase 2 of #5918
bug-ops/zeph#6234 · opened on Jul 13, 2026
1
parity: zeph lacks per-turn self-evaluation-against-goal outside autonomous sessions (Goose /goal pattern)
bug-ops/zeph#6009 · opened on Jul 11, 2026
1
research(llm): no OpenAI Batch API / Anthropic Message Batches API support for cost-tier background LLM calls
bug-ops/zeph#5929 · opened on Jul 10, 2026
1
research(vault): age identity has no hardware-backed/OS-keychain option and no key rotation flow
bug-ops/zeph#5922 · opened on Jul 10, 2026
1