agent-wiki-tasks

A tool for comparing multiple agent sessions that attempted similar tasks. It groups sessions into task families and records the results and patterns across trials or conditions.

In plain words
What is it for?
Use it to discover recurring task types, create comparison pages, and describe findings across sessions.
Why use it?
It makes repeated experiments easier to compare than reading every session summary separately.

Skill for Claude CodeCodex

Install

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.

agentmods
npx agentmods add skills/agenttoolkit/altk-evolve/agent-wiki-tasks
Any agent
npx skills add AgentToolkit/altk-evolve --skill agent-wiki-tasks
Clone the repo
git clone --depth 1 https://github.com/AgentToolkit/altk-evolve

Made for: Claude Code, Codex.

Per session 42 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,403 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00042 $0.02403
Opus 5 $0.00021 $0.01202
Sonnet 5 $0.00008 $0.00481
Haiku 4.5 $0.00004 $0.00240

Measured 3d ago against content hash ddaf8d404a70, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agent-wiki-tasks 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 3d 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.

explorations/agent-wiki/skills/agent-wiki-tasks/SKILL.md · 227 lines

How it starts

The opening of the file, as written. The whole thing — 227 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Agent Wiki — Task Comparisons

Overview

Two cognitive moves in one pass:

  1. Discover — read across all summaries and identify task families (groups of sessions that attempted the same thing across trials and conditions).
  2. Compare — for each family, write a tasks/<slug>__task.md page with a per-trial table and a findings narrative that calls out the experimental signal.

This is the cross-trajectory analysis pass of the agent-wiki family.

When to run

  • After enough summaries exist that a comparative pattern is visible (typically ≥3 sessions per family).
  • When the experiment design (e.g. trial × condition matrices) explicitly cries out for a comparison page.

Workflow

Step 1: Read the corpus

uv run python explorations/agent-wiki/skills/scripts/build_agent_wiki.py dump-summaries > /tmp/summaries.json

Output is a JSON array of one row per summary: {session_id, goal, family, trial, condition, tool_calls, errors, recall_used, summary_filename}. family, trial, condition come from existing classification rules — they may be null if no rule has matched yet.

Read the file:

Read /tmp/summaries.json

Step 2: Decide task families

For each candidate task family:

  • Slug: kebab-case identifier (e.g. extract-focal-length).
  • Family: short label used to group sessions (often equals slug, but can be looser e.g. focal-length for a slug extract-focal-length).
  • Family-match rules: how a future session gets classified. Currently supported: goal_substring: [list of substrings]. A session matches the family if its goal contains any substring (case-insensitive).
  • Tags: a few short tags.
  • Intro: 1–2 sentences setting up the question.
  • Findings: 2–5 bullets summarizing what the data shows. This is the actual product — a comparison page without findings is just a table.

Rules:

  1. A family needs ≥3 sessions. Smaller groups should not get their own page.
  2. Findings must be evidence-grounded. Cite tool-call counts, error counts, recall-used Y/N from the dump.
  3. Don't repeat what's in the table. Findings should explain why the metrics differ, not restate them.
  4. Use overrides for sessions whose goal doesn't auto-match. The override key in _config.yaml/session_family_overrides is the session id.

Read the full file on GitHub · 227 lines

Changes

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.

  1. 3d ago First seen · 227 lines · 42 tokens per session scan A ddaf8d404a70

Subscribe to this mod's changes

agent-wiki-tasks is a skill published in the GitHub repository AgentToolkit/altk-evolve (105 stars, last pushed 8d ago), licensed Apache-2.0. It adds 42 tokens to every session and 2,403 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.