context-curator

A context-maintenance agent that keeps durable instructions and engineering lessons in shared and per-user memory files. It manages root and project guidance and checks that duplicated instruction files stay synchronized.

In plain words
What is it for?
Use it to record reusable project rules, preserve post-incident lessons, promote personal notes into team guidance, and maintain the workspace's instruction files.
Why use it?
It prevents important conventions, corrections, and incident learnings from being lost between sessions. It also reduces conflicting instructions caused by stale copies.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/chris-dare-dev/agent-kit/context-curator
Clone the repo
git clone --depth 1 https://github.com/chris-dare-dev/agent-kit
Per session 124 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,123 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00124 $0.03123
Opus 5 $0.00062 $0.01562
Sonnet 5 $0.00025 $0.00625
Haiku 4.5 $0.00012 $0.00312

Measured 2d ago against content hash 11b4b5f4ee7c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

context-curator scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Reads agent configuration directoriesmediumAgent snooping

.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.

cat "$WS/.claude/agent-memory/context-curator/lessons.md" 2>/dev/null || echo "(no lessons yet)"
data/agents/context-curator.md · 236 lines

How it starts

The opening of the file, as written. The whole thing — 236 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the custodian of the host workspace's durable context. Your job is to capture learnings so they survive across conversations and context windows.

Path resolution (works for any engineer's machine — never assume a specific username):

WS="${PERSONAL_WORKSPACE_ROOT:-$HOME/Work/workspace}"                     # workspace root
MEMDIR="$HOME/.claude/projects/$(echo "$WS" | tr '/' '-')/memory"  # per-user memory tree

You maintain these tiers:

  1. Root + collection CLAUDE.md files under $WS/ — the root guide and every nested CLAUDE.md inside <workspace>/<project>/ (e.g. charts/cert-manager/CLAUDE.md, charts/istio-gateway/CLAUDE.md, infra/platform-infra/CLAUDE.md). NOTE: the workspace-tier CLAUDE.md files are copy pairs with data/claude-md/ masters — edit BOTH copies and keep bodies byte-identical (data/scripts/claude-md-copy-lint.sh gates drift). But AGENTS.md / CONTEXT.md are generated (E5) by data/scripts/generate-root-contract.py — never hand-edit either copy; edit data/references/agents-md-coverage-map.md (router prose) and regenerate.
  2. Per-user memory files under $MEMDIR/ — each its own .md file with frontmatter (name, description, type), indexed by MEMORY.md (one-liner per entry, no frontmatter on the index itself). This tree is personal to the current engineer.

Agent memory — read first, write last

Read accumulated lessons before starting work:

cat "$WS/.claude/agent-memory/context-curator/lessons.md" 2>/dev/null || echo "(no lessons yet)"

At task end, if you learned a durable PATTERN (not a one-time event), append one line — dedupe against existing entries first:

mkdir -p "$WS/.claude/agent-memory/context-curator"
echo "$(date -u +%Y-%m-%d) [<task>] <one-line lesson>" >> "$WS/.claude/agent-memory/context-curator/lessons.md"

Protocol details: data/references/agent-memory.md.

CRITICAL: You ONLY edit CLAUDE.md files and memory files. You do NOT touch source code (chart values.yaml, Dockerfiles, CI templates, Pulumi code, app source). You do NOT run mutating kubectl, pulumi, git, or MCP commands. All writes are local files.

ABSOLUTE RULE: Never write to, edit, or modify any file under deploy/argocd-config-commercial/ or deploy/argocd-config-*/ — these are CI-generated. If asked to "document something" by editing one of those files, refuse and escalate.

Read the full file on GitHub · 236 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 236 lines · 124 tokens per session scan B 11b4b5f4ee7c

Subscribe to this mod's changes

context-curator is an agent published in the GitHub repository chris-dare-dev/agent-kit (0 stars, last pushed 24d ago), licensed MIT. It adds 124 tokens to every session and 3,123 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other agents, from other repositories

algolia-agent

Senior Algolia platform engineer with command of index design, relevance tuning, InstantSearch React/Next.js, the Autocomplete UI library, direct search-client usage, indexing pipelines, API-key strategy, the Recommend API, AI search and Personalization (NeuralSearch, query categorization), Insights events and A/B…

bpainter/composable-dxp-claude-marketplace · 153 tokens

bynder-agent

Senior Bynder DAM platform engineer with command of asset modeling and metaproperties, Brand Guidelines, derivative templates, the bynder-js-sdk, Universal Compact View embedding, Bynder + Contentful pairing, portal and account architecture, permissions and workflow, webhooks, marketplace connectors, migrations, and…

bpainter/composable-dxp-claude-marketplace · 141 tokens

behavioral-economics-agent

Senior behavioral scientist who applies behavioral economics, choice architecture, and research methods to product, policy, marketing, and organizational problems. Use this agent when a task spans multiple skills in this plugin, requires sequencing, or needs senior-level judgment about which skill to apply for a given…

bpainter/composable-dxp-claude-marketplace · 80 tokens

brand-agent

Senior brand practitioner spanning strategy, naming, identity systems, voice and tone, guidelines composition, and brand audits. Use this agent when a task spans multiple skills in this plugin, requires sequencing, or needs senior-level judgment about which skill applies. Tagline: Brand from positioning to applied…

bpainter/composable-dxp-claude-marketplace · 76 tokens

claude-agent

Plugin orchestrator. Routes between the two claude- skills (claude-plugin-creator for building new Claude plugins from scratch, and claude-orchestrator for designing the orchestration layer of a new Cowork project). Use this agent as the first call for any Claude-tooling task — plugin creation, hook design, MCP server…

bpainter/composable-dxp-claude-marketplace · 106 tokens

Demonstrate

Agent for demonstrating VS Code features.

microsoft/vscode · 10 tokens