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/choiyounggi/dev-loop/knowledge-flushnpx skills add choiyounggi/dev-loop --skill knowledge-flushgit clone --depth 1 https://github.com/choiyounggi/dev-loopWhat 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 | $0.00117 | $0.03249 |
| Opus 5 | $0.00059 | $0.01625 |
| Sonnet 5 | $0.00023 | $0.00650 |
| Haiku 4.5 | $0.00012 | $0.00325 |
Grade A, and why
knowledge-flush 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 — 231 lines — stays where its author put it; the contents beside it link to each section on GitHub.
knowledge-flush — queued insights → verified wiki PR
Turn the ★ Insight candidates harvested from your sessions into a wiki
contribution, PR-only — you (the repo owner) review each PR and merge or
reject it. This skill NEVER auto-merges and NEVER pushes to main.
The queue lives at ~/.dev-loop/queue/*.jsonl (written by the Stop hook). Each
row is a candidate: trigger, directive, why, evidence, domain, tags, content.
Non-negotiable order (a PreToolUse gate enforces it)
hooks/pre-flush-pr-gate.sh blocks gh pr create on a knowledge branch unless
an INGEST_REPORT.md with three filled sections exists. So do the work first:
-
Acquire the shared flush lock — before anything else, including the checkout reset in step 1. This skill and the
hooks/auto-flush.shStop hook are the two entry points that drain the same queue; both serialize through the same lock (scripts/flush-lock.sh, issue #77).Generate this run's id once, here, and record it — every later
flush-lock.shcall in this flush (the abort-path release below, and step 5's release) must be prefixed with it. This is mandatory, not a style choice: each step in this skill runs in its own fresh shell, so nothing carries between them on its own — an unprefixed call computes a brand-new run id that no longer matches the one that acquired the lock, soreleaseis refused as a foreign caller and the lock leaks for the full TTL.RUNID="flush-$(date +%Y%m%d-%H%M%S)-$$" echo "$RUNID" # record this — every later flush-lock.sh call needs it DEV_LOOP_FLUSH_RUN_ID="$RUNID" sh "${CLAUDE_PLUGIN_ROOT}/scripts/flush-lock.sh" acquire- Exit 0 → proceed to step 1.
- Non-zero exit → do not touch
~/.dev-loop/repo. Read the failure line (held <holder-runid> <age>son stderr) and tell the user: "a flush is already running (holder<holder-runid>, started<age>sago) — stopping." Then stop. Do not retry, poll, or wait for the lock. The lock is released once, at the very end of step 5 on a successful flush. If you abort or hit an unrecoverable error at any point after acquiring it, release it before stopping — prefixed with the same$RUNIDrecorded above — so the next run does not wait out the TTL:
DEV_LOOP_FLUSH_RUN_ID="<the run id you recorded above>" \ sh "${CLAUDE_PLUGIN_ROOT}/scripts/flush-lock.sh" release
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 First seen · 231 lines · 117 tokens per session scan A 2c0ff8aae056
knowledge-flush is a skill published in the GitHub repository choiyounggi/dev-loop (4 stars, last pushed 2d ago), licensed MIT. It adds 117 tokens to every session and 3,249 once invoked, about $0.0006 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
research-first-development
Build knowledge bases that build software — research before code, teach before execute.
meditation
Consolidate session learning into permanent architecture — extract patterns into skills, instructions, prompts, or memory.
daily-log
Use when recording work sessions, tracking decisions and outcomes, or documenting lessons learned.
layered-first-principles-teaching
Use when explaining complex concepts to others, designing training materials, or preparing technical presentations with progressive disclosure.
a-evolve
Apply A-Evolve's agentic evolution methodology to improve AI agent performance across runs. Use when the user wants to diagnose agent failures, generate targeted skills from error patterns, evolve system prompts, or accumulate episodic knowledge. Works standalone or inside AutoResearchClaw pipelines. Triggers on…
pytorch-training
Best practices for building robust PyTorch training loops. Use when generating or reviewing ML training code.