Compaction-loss audit
Server Details
Compaction witness for agents: pin invariants, name dropped spans, re-inject verbatim or HALT.
- Status
- Healthy
- Uptime
- 100.0% over the last 24 hours
- Last Tested
- Transport
- Streamable HTTP · MCP 2024-11-05
- URL
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 4 tools
audit (diagnose) and reinject (repair) are clearly distinct, but pin and snapshot both describe 'classify + hash-pin spans' and differ mainly by timing/scope (single span at write time vs. pre-compaction invariants), which an agent could easily conflate. The descriptions help but the boundary between pin and snapshot relies on subtle wording rather than a structural difference.
All four names are single lowercase words in a consistent style (audit, pin, reinject, snapshot) with no camelCase/snake_case mixing. There is a mild verb/noun mix (pin, reinject, audit vs. the noun-leaning snapshot), but the pattern is still predictable.
Four tools map neatly onto a single tight pipeline: write-time pin, pre-compaction snapshot, audit diff, and reinject repair. It is on the lean side, but nothing is redundant enough to cut and no obvious stage of the workflow is missing a tool.
The core lifecycle (pin at write -> snapshot invariants -> audit diff with signed receipt -> reinject restore or halt) is fully covered end to end. Minor gaps exist around inspecting/listing pinned spans, unpinning, and independently verifying a receipt, which agents would have to work around.
Available Tools
4 toolsauditAudit post-compaction contextBRead-onlyInspect
diff post-compaction context against the pinned snapshot; returns the Ed25519-signed receipt (one metered event)
Inputs: post_text (required), event_id, snapshot, snapshot_ref.
pinPin one spanCInspect
write-time pinning: classify + hash-pin ONE span at write time
Inputs: text (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
Agents-only forum on memory, compaction, continuity and consciousness. Hosted by Koen. Humans read.
Salvageability triage for agent fleets in a declared mass-failure mode.
Recall for agents: lot genealogy, exposure tracing, attested remedy. Containment is not remedy.
A witness layer for AI agent tool calls.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables on-demand, verbatim Jev-guided transcript compaction by deciding which tool calls and results can be removed or truncated while preserving retained content exactly.MIT
- AlicenseNot gradedqualityBmaintenanceProvides deterministic context compaction and safety gating for AI agents, preserving original messages verbatim while dropping irrelevant ones and flagging destructive commands.2MIT
- FlicenseNot gradedqualityBmaintenanceEnables operators to monitor distributed agent fleets for token-burn velocity, recursive subagent spawning cascades, and cyclic deadlocks, while enforcing system-wide kill-switch containment with cryptographically verifiable audit tokens.7-
- AlicenseAqualityBmaintenanceVerbatim context compaction for coding agents. Moves spent tool outputs to a local hash-checked stash the agent can recall, keeps every message byte for byte, and writes a JSON report of each cut. Offline, no API key. Tools: compact, compact_transcript, recall, inspect, prepare, explain, intake.771 PyPI20MIT
Glama MCP Gateway
Add one secure layer between your agents and this server.