Borrowing it
Nothing to install: this file belongs to zkysar1/Claude-Mind. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/zkysar1/Claude-Mind/main/.claude/skills/reflect-maintain/SKILL.mdgit clone --depth 1 https://github.com/zkysar1/Claude-MindWrote 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/zkysar1/claude-mind/reflect-maintain)<a href="https://agentmods.dev/skills/zkysar1/claude-mind/reflect-maintain"><img src="https://agentmods.dev/badge/skills/zkysar1/claude-mind/reflect-maintain/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/zkysar1/claude-mind/reflect-maintain"><img src="https://agentmods.dev/badge/skills/zkysar1/claude-mind/reflect-maintain.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00101 | $0.08732 |
| Opus 5 | $0.00051 | $0.04366 |
| Sonnet 5 | $0.00020 | $0.01746 |
| Haiku 4.5 | $0.00010 | $0.00873 |
Grade A, and why
reflect-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 yesterday.
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 — 705 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/reflect-maintain — Maintenance Reflection
This sub-skill implements all maintenance reflection modes for /reflect. It is invoked
by the parent router for two modes:
- Memory curation mode (
--curate-memory): Retire stale/low-utilization artifacts + active forgetting - Aspiration grooming mode (
--curate-aspirations): Detect stuck goals whose evidence has converged
Each mode section below is self-contained with its own step numbering.
Mode: Curate Memory (--curate-memory)
This sub-skill implements Mode 5 of /reflect plus the Active Forgetting & Knowledge Maintenance system. It is invoked by the parent /reflect router when --curate-memory is specified, or during --full-cycle as a light sweep scoped to categories touched in the session. It retires stale/low-utilization strategies, guardrails, reasoning bank entries, and pattern signatures, and implements hippocampal-inspired active forgetting.
Triggered by: spark question sq-c03/sq-c04, --full-cycle light sweep, or stale_strategy evolution trigger.
Step 0: Load Conventions
Step 0: Load Conventions — Bash: load-conventions.sh with each name from the conventions: front matter. Read only the paths returned (files not yet in context). If output is empty, all conventions already loaded — proceed to next step.
Step 1: Gather Candidates
Scope: If triggered by spark/stale_strategy, scope to that category only. If triggered by --full-cycle, scope to categories touched this session.
1a: Stale Strategies
Bash: world-cat.sh knowledge/strategies/extracted-strategies.md
For each strategy with status: active:
stale = (today - last_reinforced).days > 30 AND times_applied == 0
If stale: add to candidates with reason "unused for 30+ days"
1b: Low-Utilization Guardrails
# rb-245 pre-read: verify counter fields exist before aggregating. Exit 1
# means the pseudocode field path has drifted from the live schema — SKIP
# this sub-phase only (other phases continue), log the mismatch for
# investigation. Do NOT --override: that silences the signal the gate exists
# to produce. Fix the field paths in the pseudocode below instead.
Bash: source core/scripts/_paths.sh && bash core/scripts/audit-schema-gate.sh \
--jsonl-path "$WORLD_DIR/guardrails.jsonl" \
--field-names "utilization.times_helpful,utilization.times_cited,utilization.retrieval_count"
# WHOLE-STORE READ BY DESIGN (g-115-3570) — NOT an unbounded-retrieval defect.
# A RETIREMENT sweep must see every active guardrail: anything outside a bounded
# top-K would be structurally exempt from retirement forever. Do NOT re-route this
# through retrieve.sh; its caps (SUPPLEMENTARY_CAPS 20/40/80) are for RELEVANCE
# ranking at decision time, which is the opposite of a corpus sweep.
Bash: guardrails-read.sh --active
For each guardrail with status: active:
# Value-density criterion (guard-841) -- NOT bare v1 utilization_score (<0.20).
# v1 ignores times_helpful/times_cited and is pre-2026-04-23 bulk-load-inflated,
# over-flagging ~75% of active guardrails (rb-2165). times_helpful and times_cited
# increment on EXPLICIT use only; times_inferred_helpful is the AUTOMATIC backstop
# that increments on retrieval-application via utilization-feedback --infer. The
# rest of the system already counts it (utility_ratio = (th + 0.5*tih)/rc), so the
# retire bar MUST include it too -- else heavily-retrieved-but-only-inferred-helpful
# entries (e.g. rb-200 tih=8) mass-retire as false-positive dead (g-115-1605).
dead = (utilization.times_helpful + utilization.times_cited + utilization.times_inferred_helpful) == 0 AND utilization.retrieval_count >= 200 AND age_days >= 60
# guard-707 gate (apply in Step 2 BEFORE retiring ANY candidate): grep CLAUDE.md +
# .claude/skills + .claude/rules + core/config for the entry ID, EXCLUDING .history/ --
# KEEP entries cited by ID in live framework files regardless of counters (load-bearing).
# Canonical bulk tool for this criterion: core/scripts/bulk-retire-dead-entries.py.
If dead: add to candidates with reason "zero value-density (helpful+cited+inferred==0) after 200+ retrievals and 60+ days"
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.
- yesterday Changed · +7 lines 69556ddd405e
- 9d ago First seen · 698 lines · 101 tokens per session scan A acc10e32f317
reflect-maintain is a skill published in the GitHub repository zkysar1/Claude-Mind (5 stars, last pushed yesterday), licensed MIT. It adds 101 tokens to every session and 8,732 once invoked, about $0.0005 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
hive.context-preservation
Proactively extract critical values from tool results into working notes before automatic context pruning destroys them.
hive.note-taking
Maintain a free-form scratchpad of decisions, extracted values, and open questions so context pruning doesn't lose anything you still need.
Vizra ADK Memory System
Implement persistent memory, session context, and vector memory (RAG) for AI agents.
learning-capture
Log corrections, errors, and discoveries to orbital/LESSONS.md so future sessions avoid repeating mistakes.
obsidian
A skill for working with an Obsidian vault, which is a folder of Markdown notes used as a personal knowledge base. It covers creating, extending, searching, and organizing notes, including links, tags, folders, and daily entries.
memory-system
3-tier hierarchical memory management for AI agents. L1 (daily active context, 100 lines), L2 (weekly decisions/insights, 400 lines), L3 (long-term archive, unlimited). Includes self-improvement loop, session continuity, semantic search, and mistake-tracking. Use when starting a session (load memory), ending a session…