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/angad-kandhari/deliberate/memorynpx skills add angad-kandhari/deliberate --skill memorygit clone --depth 1 https://github.com/angad-kandhari/deliberateWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/angad-kandhari/deliberate/memory)<a href="https://agentmods.dev/skills/angad-kandhari/deliberate/memory"><img src="https://agentmods.dev/badge/skills/angad-kandhari/deliberate/memory.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00087 | $0.01536 |
| Opus 5 | $0.00044 | $0.00768 |
| Sonnet 5 | $0.00017 | $0.00307 |
| Haiku 4.5 | $0.00009 | $0.00154 |
Grade A, and why
memory 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 6d 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 — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Memory
Memory skill for LLM coding agents. Load this when the agent keeps persistent memory across sessions — CLAUDE.md, AGENTS.md, auto-memory, a memory-bank directory, or any notes file a future session will read.
Where deliberate keeps code honest, this skill keeps what the agent remembers honest. Counters the failure modes of agent memory: bloat that buries the signal, stale entries acted on confidently, contradictory duplicates, and notes that restate what the repo already records.
How memory is stored and loaded is the harness's job. What gets written, kept, believed, and deleted is judgment — this skill governs that. The evidence is blunt: agents that record everything perform far worse than agents that record little and prune. Curation beats architecture.
1. Write Selectively
A memory earns its place by changing future behavior. Most things don't.
Before writing, both must hold:
- It will change what a future session does. A preference, a constraint, a correction, a decision with a why. Not narration of what happened.
- It can't be re-derived. If the next session could learn it from the repo, the docs, or one obvious command, writing it down is duplication, not memory.
The strongest write signal is repetition: the same mistake twice, the same correction twice, the same question twice. Once is an event; twice is a pattern worth a note.
Test: Will a future session act differently because this entry exists?
2. Never Record What the Repo Already Knows
The repo is the source of truth. A memory that copies it is guaranteed to drift from it.
Don't store:
- Code structure, file locations, function signatures — the file tree and search answer these fresh and correctly every time.
- Dependency lists, versions, scripts — the manifest already says.
- What changed and when — git history says, with proof.
- Anything a README, CLAUDE.md, or doc already states.
Store instead what the repo can't say: why a decision went this way, which approach was tried and rejected, what the user corrected, constraints that live outside the code.
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.
- 6d ago First seen · 133 lines · 87 tokens per session scan A a8fd13ee771b
memory is a skill published in the GitHub repository angad-kandhari/deliberate (2 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 87 tokens to every session and 1,536 once invoked, about $0.0004 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
session-handoff
Create consistent session handoff documents for context transfer between sessions. Use when closing a session, reaching 75% token consumption, or when user requests handoff. Ensures proper YAML headers for metrics collection, LF line endings, and complete documentation of progress, decisions, and next steps.
context-scaffold
Analyzes an existing project and generates context files (AGENTS.md, context directory, cascading structure) pre-populated with discovered information. Run once to bootstrap, then customize.
rk-pull
Pull knowledge base from remote and merge into local. Fetches remote state, Agent compares diffs and merges (alias LRU merge for index.md, union dedup for registry.md, newer-cached-date wins for docs). Does NOT push — local only. Use when user says 'pull knowledge', 'sync from remote', 'rk-pull', or on a new machine…
rk-push
Push local knowledge base to the remote repository. Always fetches remote first, Agent compares diffs and merges (alias LRU merge for index.md, union dedup for registry.md, newer-cached-date wins for docs), then commits and pushes. Retries once on push rejection. Use when user says 'push knowledge', 'sync to remote'…
rk-update
Incrementally update a codebase knowledge base by checking git commit history. Finds changed files since last update, regenerates only affected documentation, and updates the index. Auto-pulls from remote before updating and auto-pushes after if remote is configured. Use when user says 'update knowledge base'…
rk-memo
Save a piece of personal knowledge to the user's personal knowledge base (personal project). No git repo needed — stores anything the user knows by heart: commands, tips, best practices, personal workflows. Searchable via rk-search, synced via rk-push/pull. Use when user says 'remember this', 'save this note'…