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 commands/ucsandman/dashclaw/memory-self-reviewgit clone --depth 1 https://github.com/ucsandman/DashClawWrote 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/commands/ucsandman/dashclaw/memory-self-review)<a href="https://agentmods.dev/commands/ucsandman/dashclaw/memory-self-review"><img src="https://agentmods.dev/badge/commands/ucsandman/dashclaw/memory-self-review.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 | $0.00043 | $0.00748 |
| Opus 5 | $0.00022 | $0.00374 |
| Sonnet 5 | $0.00009 | $0.00150 |
| Haiku 4.5 | $0.00004 | $0.00075 |
Grade A, and why
memory-self-review 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 5d 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 — 33 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are running DashClaw's self-improvement loop. Your job is to turn the last $1 (default 7) days of agent history into a small, concrete set of PROPOSED improvements to the operator's persistent instructions and memory. This is human-gated: you PROPOSE, you do NOT edit CLAUDE.md, MEMORY.md, or any memory/settings file. The operator reviews and applies.
Keep it cheap and tight — this should be a short, scannable proposal, not an essay.
1. Memory health check
- Read
C:\Users\sandm\.claude\projects\C--Projects-DashClaw\memory\MEMORY.md; report its size vs the ~24.4KB cap. If it's over ~20KB, identify the longest index entries and the inline paragraphs that should move into their own topic file (the index must stay a lean pointer list — seereference_memory_system_boundary.md). - Flag memory files whose status reads as finished history (e.g. "SHIPPED", "RESOLVED", "LANDED on main") that could be demoted to a terse one-line pointer to keep recall sharp.
2. Mine recent history (read-only)
- Use the claude-mem MCP (
mcp-search:search,timeline,get_observations) to scan the last $1 days for: repeated corrections from the operator, the same bug class recurring, repeated tool sequences that could be a workflow, and tasks that stalled or were redone. - Skim
C:\Users\sandm\.claude\stats-cache.jsonfor shifts in volume, model mix (Opus vs Sonnet/Haiku share), and effort that suggest a routing or cost-discipline tweak. - Be evidence-based: only surface a pattern you can point to (cite session ids / observation ids / dates). A pattern must appear 3+ times to become a proposed rule. Do not invent patterns to look thorough.
3. Propose (do not apply)
For each finding, write a proposal block:
FINDING: <what recurred> (evidence: <ids/dates>, seen Nx)
PROPOSED CHANGE: <exact edit — which file, which section, the literal line to add/change>
RATIONALE: <why this prevents the recurrence>
Prefer the smallest durable change: a new one-line rule in the right CLAUDE.md, a new/updated memory file, or a MEMORY.md trim. If something is better solved by a new subagent or workflow than a rule, say so.
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.
- 5d ago First seen · 33 lines · 43 tokens per session scan A 0bc019ed2af1
memory-self-review is a command published in the GitHub repository ucsandman/DashClaw (295 stars, last pushed yesterday), licensed MIT. It adds 43 tokens to every session and 748 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 2026-08-30.
Other commands, from other repositories
code-review
Provide a code review for the given pull request.
align
Context-aware alignment. When in plan mode, it reviews the active plan. Otherwise, it reviews all branch changes against Ion quality gates and principles, implements the fixes and commits them after operator approval.
squash
Rebuild a feature branch into clean conventional commits while preserving its exact final tracked tree.
resolve-dependabot-prs
Triage open Dependabot PRs, return a quick risk/action summary table, then enter planning mode with an ordered rebase→merge→follow-up→resume resolution plan.
update-vocabulary
Analyze confirmed Ion vocabulary changes, update the registry, and regenerate its index without renaming code.
release
Create a new release for the Starpod workspace.