Files
msd-core/docs/context-monitor.md
Tom Boucher 334a64168e chore(npm): rebrand packages to @opengsd scope (#127)
* chore(npm): rebrand packages to @opengsd scope

Rename:
- get-shit-done-redux → @opengsd/get-shit-done-redux
- @gsd-redux/sdk → @opengsd/gsd-sdk

Add publishConfig.access=public for first-time scoped publish.
CLI binary names (get-shit-done-redux, gsd-sdk, gsd-tools) unchanged.

Sweeps install commands, npx invocations, CI publish/version-check
workflows, tests, docs, READMEs (all translations), and the
PACKAGE_NAME constant in check-latest-version.

Bumps qs 6.15.1 → 6.15.2 to clear a moderate advisory surfaced by
the audit-clean test (GHSA-q8mj-m7cp-5q26).

Closes #126

* chore: pin 2.0.0 release + remove canary workflow

- Bump both packages 1.50.0-canary.0 → 2.0.0 for first @opengsd publish
- Remove .github/workflows/canary.yml and canary dist-tag handling in
  release.yml / release-sdk.yml
- Drop canary section from VERSIONING.md

Refs #126

* chore: address review findings + harden tarball-smoke timeout

- .changeset/opengsd-org-rename.md: match project's custom
  parse.cjs frontmatter (type: Changed / pr: 127); the scoped
  @changesets/cli keys were silently rejected.
- CONTEXT.md: drop two canary-stream policy lines and a dangling
  DEFECT.CANARY-VERSION-LEAK.cross-ref now that canary.yml is gone.
- tests/release-tarball-smoke.install.test.cjs: pass
  timeout: 600_000 for npm pack + global install; the 3-minute
  runNpm default was timing out on slower Docker hosts (cartographer).

Refs #126

* fix(sdk): add missing type/runtime devDependencies for build

prepublishOnly invokes tsc which couldn't resolve @types/node,
@types/ws, or synckit. They had been hoisted from root but were
not declared in sdk/'s own package.json — first publish from a
clean SDK tree failed.

Refs #126

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ci): use npm pack stdout instead of glob to find tarball

`npm pack --silent` for a scoped package (@opengsd/get-shit-done-redux)
produces `opengsd-get-shit-done-redux-*.tgz`, not `get-shit-done-redux-*.tgz`.
Capture the filename from stdout instead of a hardcoded glob so the step
works regardless of package name format.

Fixes smoke (ubuntu-latest, 22, false) CI failure.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* ci: treat workflow-file changes as test-skip eligible

`.github/workflows/install-smoke.yml` (and other workflow files)
were in neither `test.yml` paths nor `test-skip.yml` paths-ignore,
so neither workflow ran on a workflow-only commit — leaving the
required test-skip check perpetually missing.

Refs #126

* chore: reset version to 1.0.0 for first @opengsd publish

Nothing has been published yet under the @opengsd scope, so the
inaugural release uses 1.0.0 rather than 2.0.0. The "major bump"
in the changeset reflects the breaking install-command change for
users migrating from the prior unscoped `get-shit-done-redux`, not
a numeric continuation from a 1.x line under the new identity.

Refs #126

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-22 16:22:41 -04:00

3.3 KiB

Context Window Monitor

A post-tool hook (PostToolUse for Claude Code, AfterTool for Gemini CLI) that warns the agent when context window usage is high.

Problem

The statusline shows context usage to the user, but the agent has no awareness of context limits. When context runs low, the agent continues working until it hits the wall — potentially mid-task with no state saved.

How It Works

  1. The statusline hook writes context metrics to /tmp/claude-ctx-{session_id}.json
  2. After each tool use, the context monitor reads these metrics
  3. When remaining context drops below thresholds, it injects a warning as additionalContext
  4. The agent receives the warning in its conversation and can act accordingly

Thresholds

Level Remaining Agent Behavior
Normal > 35% No warning
WARNING <= 35% Wrap up current task, avoid starting new complex work
CRITICAL <= 25% Stop immediately, save state (/gsd-pause-work)

Debounce

To avoid spamming the agent with repeated warnings:

  • First warning always fires immediately
  • Subsequent warnings require 5 tool uses between them
  • Severity escalation (WARNING -> CRITICAL) bypasses debounce

Architecture

Statusline Hook (gsd-statusline.js)
    | writes
    v
/tmp/claude-ctx-{session_id}.json
    ^ reads
    |
Context Monitor (gsd-context-monitor.js, PostToolUse/AfterTool)
    | injects
    v
additionalContext -> Agent sees warning

The bridge file is a simple JSON object:

{
  "session_id": "abc123",
  "remaining_percentage": 28.5,
  "used_pct": 71,
  "timestamp": 1708200000
}

Integration with GSD

GSD's /gsd-pause-work command saves execution state. The WARNING message suggests using it. The CRITICAL message instructs immediate state save.

Setup

Both hooks are automatically registered during npx @opengsd/get-shit-done-redux installation:

  • Statusline (writes bridge file): Registered as statusLine in settings.json
  • Context Monitor (reads bridge file): Registered as PostToolUse hook in settings.json (AfterTool for Gemini)

Manual registration should use the absolute Node executable path that ran the installer. On Windows PowerShell, prefix the command with & when that executable path is quoted.

Manual registration in ~/.claude/settings.json (Claude Code):

{
  "statusLine": {
    "type": "command",
    "command": "\"/usr/local/bin/node\" \"/Users/me/.claude/hooks/gsd-statusline.js\""
  },
  "hooks": {
    "PostToolUse": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "\"/usr/local/bin/node\" \"/Users/me/.claude/hooks/gsd-context-monitor.js\""
          }
        ]
      }
    ]
  }
}

For Gemini CLI (~/.gemini/settings.json), use AfterTool instead of PostToolUse:

{
  "hooks": {
    "AfterTool": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "& \"C:/Program Files/nodejs/node.exe\" \"C:/Users/me/.gemini/hooks/gsd-context-monitor.js\""
          }
        ]
      }
    ]
  }
}

Safety

  • The hook wraps everything in try/catch and exits silently on error
  • It never blocks tool execution — a broken monitor should not break the agent's workflow
  • Stale metrics (older than 60s) are ignored
  • Missing bridge files are handled gracefully (subagents, fresh sessions)