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 skills add waqas1412/claude-harness --skill harness-distillgit clone --depth 1 https://github.com/waqas1412/claude-harnessWrote 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/waqas1412/claude-harness/harness-distill)<a href="https://agentmods.dev/skills/waqas1412/claude-harness/harness-distill"><img src="https://agentmods.dev/badge/skills/waqas1412/claude-harness/harness-distill/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/waqas1412/claude-harness/harness-distill"><img src="https://agentmods.dev/badge/skills/waqas1412/claude-harness/harness-distill.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00117 | $0.01262 |
| Opus 5 | $0.00059 | $0.00631 |
| Sonnet 5 | $0.00023 | $0.00252 |
| Haiku 4.5 | $0.00012 | $0.00126 |
Grade A, and why
harness-distill 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 4d 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 — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/harness-distill: mine sessions into verified learning proposals
A recipe, not an engine. It closes the harness's one-directional gap: learnings from real sessions
are otherwise captured only by hand. It carries no Write/Edit tool, so it is structurally
propose-only: the brain plans and delegates, read-only advisors fan out, and every survivor is
emitted as a proposal for your approval, never a silent mutation. On-demand only; there is no
scheduler. Invoking this skill by name is the explicit in-the-moment opt-in the global single-main-loop rule requires; it licenses the fan-out described below and nothing else. The verbose rubric, skeptic prompt, dedup rule, and proposal template live in
references/distill-pipeline.md, opened when a step needs them.
Scope discipline
This is a prose choreography over native subagent dispatch, like /orchestrate. Do not build a
coordinator, daemon, message bus, or scheduler. Default to a tight bounded slice (the current
session plus a small recent window for this repo) because mining transcripts is token-heavy; the
user can widen via the argument. Use effort, not model tiers: low effort for batch extraction and
clustering, high effort for the skeptic pass. Never pin a per-agent model.
Step 1: Scope and gather (read-only)
Default slice: the current session plus a small recent window of this repo's transcripts at
~/.claude/projects/<slug>/*.jsonl, plus code-review corrections (prior /code-review output and
gh pr review comments). If those transcripts are absent (a fresh install), say so and stop; there
is nothing to distill. Extract candidate signals with Bash plus jq over the slice; reserve a
mechanical-tier haiku agent per session only for a large batch. Candidate signals: explicit user
corrections, repeated friction, gotchas hit, and successful ad hoc approaches. See the extraction
heuristics in references/distill-pipeline.md.
Step 2: Cluster (sonnet, parallel)
Fan out agents to group the raw candidates by theme in one pass. Recurrence across two or more sessions is the signal; a one-off is noise and drops out here.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 4d ago Changed 52c1186844ef
- 12d ago First seen · 84 lines · 117 tokens per session scan A ee60546cc002
harness-distill is a skill published in the GitHub repository waqas1412/claude-harness (1 stars, last pushed 5d ago), licensed MIT. It adds 117 tokens to every session and 1,262 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
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.
caveman-compress
Compress natural language memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserves all technical substance, code, URLs, and structure. Compressed version overwrites the original file. Human-readable backup saved as FILE.original.md. Trigger: /caveman-compress FILEPATH or…
continuous-learning-v2
Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents.
lat-md-knowledge-graph
Design or audit a repo-local markdown knowledge graph with wiki links, source-code backlinks, drift checks, and searchable sections. Use when AGENTS.md/CLAUDE.md is too flat for a large codebase or when a custom harness needs durable structured project memory.
flame-harness-evaluator
Phase 6 — skeptical QA. Run the game, watch it, then judge against the contract. Default = functional check; --strict adds quality and edge-case passes.