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.
npx agentmods add skills/d2a8k3u/claude-code-memory/memory-maintainnpx skills add d2a8k3u/claude-code-memory --skill memory-maintaingit clone --depth 1 https://github.com/d2a8k3u/claude-code-memoryWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00040 | $0.01726 |
| Opus 5 | $0.00020 | $0.00863 |
| Sonnet 5 | $0.00008 | $0.00345 |
| Haiku 4.5 | $0.00004 | $0.00173 |
Grade A, and why
memory-maintain scanned grade A with 0 findings 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 131 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Memory Maintenance
Clean up noise, deduplicate records, and reorganize large memories. Run this after upgrading the plugin, when search quality degrades, or when memories have grown large and unfocused.
Steps
1. Assess
Run memory_health to get current counts, types, and staleness stats.
2. Deduplicate accumulative records
These records should exist at most once each. Search for duplicates and merge them.
Tech-stack:
memory_search("Technology stack used", type="semantic")- If multiple records have the
tech-stacktag AND content starts withTechnology stack used::- Collect all technology items from each record's comma-separated list
memory_updatethe most comprehensive record with the union of all items (sorted alphabetically)memory_deleteall other duplicates
- Tag collision check: If any record has the
tech-stacktag but does NOT start withTechnology stack used:(e.g., a detailed dependencies description from memory-init), rename its tag fromtech-stacktotech-stack-detailviamemory_updateto prevent merge conflicts.
Active modules:
memory_search("Active modules directories", type="semantic")- If multiple records have the
active-modulestag AND content starts withActive modules/directories::- Same approach: union all items, update the best, delete the rest
- Tag collision check: Same as above — rename non-accumulative records' tag to
active-modules-detail.
3. Fix episodic titles
Episodic memories created before v1.2 often have no title, causing their content (starting with **Task:** or **Files modified:**) to be used as the label. This makes them indistinguishable in graphs and lists.
memory_list(type="episodic", limit=100)— scan all episodic records- For each record that has no title (title is null/empty):
- If content starts with
**Task:**— extract the task text after the marker, strip markdown bold, and use first ~80 chars as the title - If content starts with
**Files modified:**— derive a title like"Session: N files in dir1, dir2" - If content contains both
**Task:**and**Files modified:**lines — prefer the task text for the title memory_update(id, { title: derivedTitle })
- If content starts with
- Skip records that already have a meaningful title
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.
- 2d ago First seen · 131 lines · 40 tokens per session scan A aad83847e319
memory-maintain is a skill published in the GitHub repository d2a8k3u/claude-code-memory (7 stars, last pushed 2mo ago), licensed MIT. It adds 40 tokens to every session and 1,726 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
memory-assistant
Comprehensive memory management skill that guides memory capture and recall throughout sessions. Triggers when: user makes architectural decisions ("let's use", "we should", "decided to", "choosing"), discovers learnings ("TIL", "learned that", "discovered", "found out", "realized"), encounters blockers ("stuck on"…
memory-recall
This skill should be used when the user asks about "past decisions", "previous learnings", "what did we decide", "how did we solve", "remember when", or mentions topics that may have stored memories. Also triggers when working on tasks where historical context would be valuable, such as "working on authentication"…
opencli-sitemap-author
Use when creating or maintaining OpenCLI site sitemaps: agent-facing navigation, page-state, action, workflow, API-reference, pitfall, and fallback knowledge for a website. Use after browser exploration discovers durable site context, when a sitemap is stale, or when promoting local site knowledge into the repo.
golden-rss
Use when testing the rss golden build.
omh-code-review
This is a Hermes-native code-review workflow skill.
redteam-web-detail-pack
Routing and boundary guidance for authorized general web application security testing. Use as a web testing router when the attack surface should be dispatched to more specific web vulnerability skills.