Plugins

Session-level verdict on whether compaction would lose unlanded state; structurally refuses per-block screening.

Extends the composer model selector menu with a third Context window row, preset capacities (64K–2M), custom input, provider-wide sync, and adaptive compaction guard.

Prose review for human-facing writes. The write and edit tools require a human_facing flag; the plugin reviews the text against layered PROSE.md rules in the background, stages annotations, reconciles them against the full file, and merges them through /humanize, /humanize-loop or post-compaction while the main session keeps working. /humanize-review runs a batch over git-tracked files on demand.

Notes-backed context compression for DeepSeek Harness: replaces LLM-summarized compaction with a model-written persistent notebook plus recoverable raw history, so the rollover boundary itself costs zero model calls.

Chunked compaction for oversized sessions: takes over only when the live context exceeds 85% of the model window (or the request exceeds 4 MB), summarizes the history in 1.5 MB chunks and merges them; smaller sessions keep the built-in compaction.

Context observability and control for DeepSeek Harness: cross-session token analytics, session-level budget/prompt/tool controls, and optional agent scaffolds. It reuses the official tokenMeter, compaction, systemPrompt, and tools.restrict seams; monitoring is read-only.

Indexes session transcripts into the Akasha session layer as turns end, records successful compactions, and adds a per-turn rhythm line to the runtime context. Companions: Hindsight (repo memory), OpenViking (user memory), and the built-in session search (see repo README). Setup: clone the repo, copy core/ aside, then replace <AKASHA_DIR> in cordis.patch.yml with its absolute path (or set the AKASHA_DIR env var); plugins fail loudly while the placeholder remains.

Registers /compact_queue, a compaction command that does not drop the request when the session is busy: it compacts immediately when idle, otherwise queues the request in that session and runs it at the session's next idle window, with a 2-second fallback tick beside the agent status event so the trigger is not missed, a 30-minute wait cap, and /compact_queue cancel to withdraw a queued request or abort a running one. Each session keeps its own queue state from a single command registration, completion and failure arrive as a collapsed notice that does not wake the model, and every ManualCompactionError code except busy is terminal so a failure is not retried. Host half only; the core /compact command is left untouched.

Stops compaction summaries and session titles from burning their output budget on reasoning, by lowering the thinking level and raising the token floor for auxiliary calls only.

Replaces rolling narrative summaries with an append-only session index, so early facts survive compaction verbatim and shadowed text stays retrievable by anchor.

Inspect conversation context and apply configurable compaction with usage history.

macOS only: holds a PreventUserIdleSystemSleep assertion (caffeinate -i -w <host pid>) while any DSH session runs, and releases it when the last one stops, so a long turn is not cut off by idle sleep. The display may still turn off. It keeps holding through the maintenance phase, where Agent.status reports idle and a long manual /compact runs, via the compaction/start and compaction/end events. It reconciles against ctx.agents.list() every 60s, so a session already running when the plugin is hot-reloaded is picked up at once instead of at the next turn boundary. A failed hold retries with exponential backoff capped at 60s and never gives up; release escalates from SIGTERM to SIGKILL after 2s, and -w lets caffeinate exit by itself if the host dies. No browser half — configured through the profile patch. 51 tests, including two that read pmset -g assertions on real hardware.

Uses a dedicated model for context compaction rescue and adds manual compaction to the native context popup.

Route conversation-compaction calls to a chosen summarizer model in the Web settings page, and retry them on a configured fallback model when the primary ends in an error.

Session-context surfaces ported from T3 Code: a session-row hover card showing machine, git branch, model, agent preset, approval policy and token usage, a status pill covering pending approval, awaiting input, working, error and subagent work, a provider and model chip, a per-session header strip, a lineage breadcrumb, row-menu actions that copy the context or run DSH compaction, and a preference row in DSH Settings for choosing which facts appear. Nothing is added to the composer.

Persistent memory for dsh in the style of Hermes Agent: a memory tool that keeps a user profile and the model's own notes, delivered to every chat as a logged context message — at the start, after a change and after compaction — and a session_search tool that reads earlier chats only after the user approves, with a Memory panel to see and edit what is kept.

Tiered compaction hub: replaces the single 80% compaction cliff with a configurable ladder (default 30/50/70/90) whose tiers escalate from gentle to maximum consolidation laws, folds a deterministic keyword index into every checkpoint so facts accumulate across cycles, and summarizes through a model chain — primary plus up to 3 fallbacks, each with its own reasoning effort. Adds a Compaction module sub-tab and a Settings card; toggling off always restores stock compaction.

Context compaction that keeps its receipts: a five-section handoff summary, a durable segment index, review tools that read compacted text back, optional working notes, and a read-only extraction exit. Mounts as the compaction-basic row's replacement; cha

Host- and agent-layer DeepSeek Harness plugin that prices every automatic context rewrite before it lands, asks the human by default, and reports what the provider had to re-read cold.

Rapid-refill circuit breaker for DeepSeek Harness automatic compaction. It refuses a compaction that would be futile — the context refilled within fewer than N tool turns M times in a row, which the unbounded step-pressure trigger cannot detect — and instead of burning a summarization call on every step it reports the situation with advice, usually that a single read or tool output was too large. Shows state and re-arms through /compaction-breaker, and injects one prompt section so the model relays it to the user. Covers host-plane sessions through its profile bundle; ordinary sessions need one row replaced inside an agent preset.

Model-driven context compaction for DeepSeek Harness. The agent decides when and what to compress, commits durable session-log replacements, and ships reversible tool-result compression with byte-exact hash retrieval.

Rotate the outbound LLM session ID after each successful DSH compaction without creating a new DSH session.

RP memory built on pmp tavern: RP history summaries ride the native compaction pipeline, with plain tool sessions kept out of the way. Full recall requires the downstream anima plugin. Requires the plugin-provided roleplay preset to be enabled.