contextsaver

Diagnoses wrong processes, wrong workflows and wrong habits in Claude Code sessions — and lets you fix them in one click.

Sessions bog down when the process is wrong. Review and fix rounds per lane, stacked on a whole-branch review that already covers them. The premium model on mechanical stages. Agents spawned for jobs a single command would do. The full check suite after every merge. Tool-level waste — the same file read again, the full suite after a one-line edit — compounds on top. ContextSaver catches both: the wrong process at the moments that matter, and the wrong habit as it builds. It gives you the measured cost and one click to change it.

Features

• It diagnoses the process, not just the calls. "Per-lane review rounds duplicate the whole-branch review — 3 rounds × 3 lanes, ~4h so far. One whole-branch review reaches the same result." Process findings fire at moments that matter: an accepted plan, a Workflow launch, a pace complaint, every thirty minutes.
• It names tool-level habits too. "Claude keeps running the whole suite after every one-file edit — 5×, ~9% of context, 4m each." The calls behind the claim are one keypress away.
• Nothing to configure. No thresholds, no rules to tune. The plugin gathers the evidence and the model asks the right question; that catches things nobody wrote a rule for.
• Fixes reach every subagent. When you fix a habit, the instruction rides the prompt of every new agent the session spawns. The card shows "sent ×N" so you can see it landed.
• It works mid-turn. Long agentic turns are checked while they run, and your fix reaches Claude on its next tool result, then rides every prompt after it, so it survives compaction.
• It shows where your session went. One line each for time and context, and a plain sentence on what those minutes and tokens actually bought.
• It stays quiet. A single occurrence is never a finding. A wrong card costs more than a missed one.
• Fixes outlive the session. Turn a decision into a CLAUDE.md rule, a skill, an agent brief or a permission rule, written only when you click Write.
• It never touches your work. No tool denied, no output trimmed, no error hidden. If a hook throws, your session carries on as if the plugin weren't there.

Install

[!NOTE] Needs Claude Code 2.1.273 or newer. ContextSaver is a Claude Mod, built on function hooks, which are early access — so enable the flag first.

export CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1   # your shell, or ~/.claude/settings.json under "env"

Then, in Claude Code:

/plugin marketplace add AlmogBaku/ContextSaver
/plugin install contextsaver@contextsaver

That is the whole setup: no config, no API key, no dependencies, no build step. Installing mid-session works too — the plugin reads what already happened out of the transcript and checks it right away, without waiting for your next prompt.

Usage

1. Work as usual. One line sits above your prompt and stays quiet until something repeats.
2. A card appears in the pane beside your transcript: what Claude keeps doing, how often, what it has cost you in context and minutes, and what to do instead. Press i for the calls behind it.
3. Choose. Fix sends the suggested fix, Fix… lets you reword it first, Ignore drops it for the session.
4. Claude changes course in the turn already running, and the pane credits what you saved — or tells you the instruction was ignored.
5. Keep what worked. Each decision is offered as a CLAUDE.md rule, a skill, an agent brief or a permission rule: Write saves it, Try uses it for this session only, Skip drops it.

[!TIP] The pane docks beside your transcript in a wide terminal and sits above the prompt otherwise. /saver toggles it at any width, ctrl+x tab moves the keyboard into it, Tab cycles the buttons, Enter presses, and Esc hands the keys back.

Commands

How it works

Every tool call becomes a row in a session ledger: what ran, how long it took, how much it added to context, which files it touched. The plugin runs two judges.

The process judge fires at moments that matter — an accepted plan, a Workflow launch, a pace complaint you type, every thirty minutes. It builds a digest of what the user asked, how the work was organised, what models ran which roles, how many merges got a check, what was re-read and what was slow. From that, it asks: given what the user asked, how would a lean expert run this work, and where does this session diverge? At most three findings, each with a measured cost.

The habit judge runs continuously — about every 30k tokens and three turns, or every 40 calls and five minutes inside a long turn. It asks what has repeated and bogged things down, and whether there was a shorter path. Whatever it finds becomes a card, with the costs computed from the ledger.

Fix on a process card sends a one-time re-plan to the main loop. Fix on a habit card sends a standing instruction that rides every subsequent prompt and is also appended to every new subagent's prompt — the card shows "sent ×N" to confirm it landed. Neither ever blocks a tool.

The architecture, the judge's prompt and the design brief are in docs/SPEC.md; the product spec is docs/PRD.md.

Notes and limits

[!IMPORTANT] The audit runs on your session's model, so a session on Opus pays Opus for it. It keeps itself to a few percent of the session's tokens, and /saver debug shows exactly what it spent.

• Early access. Function hooks are new and the API underneath can still change. Every module is tested and the main flow is verified live, but expect rough edges.
• Short sessions stay quiet. Under ~15 tool calls or 5 turns, only behaviour seen three or more times is reported.
• Durations are wall time. They include the time a permission prompt spent waiting for you, and the model is told as much.
• Terminal and desktop only. On mobile surfaces the plugin keeps its ledger and draws nothing.
• The band speaks for dead turns and running workflows. After a turn that ended in an API error or a refusal it reads ✕ Last turn ended in … · type anything to continue until your next prompt; while a workflow runs and nothing is found it names the run, its stage, its agents and their calls.

Development

git clone https://github.com/AlmogBaku/ContextSaver && cd ContextSaver
./scripts/check.sh                                          # validate --strict, typecheck, tests
CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 claude --plugin-dir .    # run Claude Code with this folder loaded

All logic is pure functions over a single State in hooks/core/. hooks/register.ts is the only file that touches events, and every hook falls back to next(e) on any path it does not own. hooks/ui.tsx renders two view models and never reads State. With CONTEXTSAVER_DEBUG=1, /saver demo fills the pane with sample cards, so the drawing can be worked on without waiting for a real finding.

Under --plugin-dir, editing a file hot-reloads the plugin and resets session state. Learned patterns persist in the plugin store.

License

MIT © Almog Baku. See LICENSE.