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 3xachris/3xa-harness --skill decision-loggit clone --depth 1 https://github.com/3xachris/3xa-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/3xachris/3xa-harness/decision-log)<a href="https://agentmods.dev/skills/3xachris/3xa-harness/decision-log"><img src="https://agentmods.dev/badge/skills/3xachris/3xa-harness/decision-log/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/3xachris/3xa-harness/decision-log"><img src="https://agentmods.dev/badge/skills/3xachris/3xa-harness/decision-log.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 | $0.00099 | $0.02638 |
| Opus 5 | $0.00049 | $0.01319 |
| Sonnet 5 | $0.00020 | $0.00528 |
| Haiku 4.5 | $0.00010 | $0.00264 |
Grade A, and why
decision-log 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 9d 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 — 146 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Decision Log
A fresh session inherits what was written down and nothing else. The log carries a decision across that boundary, so it is written for the cold read: someone who was not there, searching months later, with no memory to fill the gaps. What that reader needs above all is to tell, from a scan, which entry holds what they came for.
1. Shape of the log
- One file with two layers. The entry layer is append-only: once written, an entry stays as written, and a change of mind arrives as a new entry that references the old one. The summary layer is a maintained field at the top and is overwritten as the current state changes. Reading the entries top to bottom shows how the thinking moved, which is usually the question being asked.
- Every entry opens with a fenced YAML block, directly under its heading:
Four keys, always present:## 2026-08-07 Dropped the write-through cache ```yaml type: decision status: closed tags: [architecture] source: docs/backlog.md ```type— one ofdecision(a choice made),amendment(a frozen scope changed, fromworkorder),rejection(a human turning an artifact down, fromsensory-gate),correction(a human correcting the work),closeout(a task finished and reported, fromhonest-closeout). A value nothing produces is a value the format check spends its strictness guarding for no one.status—openif this entry leaves something for a later entry to close (a gate awaiting disposition, a decision flagged for revisit),closedif the entry is self-contained. Set once, at write time, like every other part of the entry — an entry that later gets closed is closed by a new entry pointing back at this one with a WikiLink (§4), never by editing this block.tags— free-form labels for grouping and filtering,[]if none apply.source— path from the project root to where the full detail lives, or the bare wordnone. A gate or a closeout entry carries extra keys beyond these four (§5) — same idea as Obsidian properties, wheretags/aliases/cssclassesare the defaults and anything else you add is still a property. This block is what makes an entry parseable by machine and skimmable by human. Write it at the time — it is the one part of an entry that cannot be reconstructed afterwards. - Not literal file frontmatter. Real YAML/Obsidian frontmatter is a file-level concept — one block, at the very top of the file. This log is one file with many entries, so what sits under each heading is a YAML-shaped fenced block, not something Obsidian's Properties panel will render specially. It is still genuine YAML: parseable with a real parser instead of an ad hoc line regex, and it carries the same four keys the frontmatter convention specifies (
type/status/tags/source) so a decision that later graduates into its own note file needs no field rework, just the---delimiters swapped in. SeeCASES.mdif you're deciding whether to split this file into one-note-per-decision instead — that trade is real, and not the one this format makes. - Entries point. Roughly five lines below the block: what was decided, the one-line reason, and where the full detail lives. When an entry starts growing, the content belongs in a real document and the entry belongs pointing at it.
- Update the summary layer — open items, current focus, counts — in the same edit as the entry below it. It gives the current state in one glance, and stays trustworthy for exactly as long as that pairing holds.
- Split at a threshold you set in advance — 200 KB unless you have a measured reason for another number. The point is having a line at all: file-reading tools truncate a long log silently, and a log you believe you read in full is worse than one you know you didn't. At the line, older entries move to a dated archive, the live file keeps the current period, and both the summary and the archive index are updated in the same pass.
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.
- 9d ago First seen · 146 lines · 99 tokens per session scan A 3fb782f0f1d2
decision-log is a skill published in the GitHub repository 3xachris/3xa-harness (17 stars, last pushed 9d ago), licensed MIT. It adds 99 tokens to every session and 2,638 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 2026-08-30.
Other skills, from other repositories
auto-resume
Recover active change and next action from artifacts. Use on fresh session with existing work.
agent-carnet
Use this skill when the user asks to save, recall, find, or organize notes. Triggers on: 'remember this', 'save this', 'note this', 'what did we discuss about...', 'check the notebook', 'find in carnet'. Also use proactively when discovering findings worth preserving across sessions.
remember
Review reusable project knowledge and decide what belongs in project memory, notepad, or durable docs.
grimoire
Use when the user says 'update context', 'update claude', 'save library', or after significant project changes.
mnemos
Task-scoped memory lifecycle — typed MnemoGraph prevents lossy context compaction by treating facts/decisions/code-refs/handoffs as distinct node types with per-type eviction policies.
ln-31-performance-optimizer
Profiles and improves a measured latency, throughput, CPU, memory, or I/O problem. Not for speculative tuning or cosmetic refactoring.