algorithm-update

Run a maintainer algorithm-update session — synthesize the collected Q2 ("smarter algorithm") reflections, cluster the recurring ideas, and fold approved changes into ALGORITHM.md. Use when the session reminder shows "Algorithm Review Due", or when asked to review/update the algorithm from reflections.

Jump to install

Repository: kovrichard/portable-agent-layer
Last source activity: May 31, 2026 at 08:04
Detected SKILL.md language: English
Stars: 12
Forks: 0

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/kovrichard/portable-agent-layer --skill algorithm-update

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer

2 files

SKILL.md

Source instructions · Read-only preview

Back to search Back to installation

Algorithm update session

This is a repo-only, maintainer workflow. It edits the algorithm source in this checkout (assets/templates/PAL/ALGORITHM.md), which is then committed and shipped to all users on the next release. It only makes sense inside the portable-agent-layer repo; it is never installed downstream.

Workflow

Synthesize the reflection backlog since the last review:

bun src/tools/agent/algorithm-synthesize.ts            # human report
bun src/tools/agent/algorithm-synthesize.ts --json     # full membership, nothing capped

Pass --since <ISO> to scope to a window if asked.

Cluster semantically — do not trust the keyword buckets. They are a coarse pre-sort biased to one model's wording. Read the report's Unbucketed section in full (that is where missed phrasings land) and the --json membership, then group by actual meaning, not shared keywords. Merge buckets that are the same idea worded differently.

Rank candidate changes by recurrence × breadth (how many distinct projects raised it). Surface the top recurring structural ideas as concrete proposed edits to ALGORITHM.md — a new gate, a phase step, a criterion rule, a capability pre-check.

Present to the user and get per-change approval. This is human-in-the-loop — never edit the algorithm unilaterally. For each approved change, edit the source:

Edit assets/templates/PAL/ALGORITHM.md (the canonical source).
Never edit ~/.pal/docs/ALGORITHM.md — it is engine-managed and overwritten on pal install.
If the change adds a visible step, also add its output token to the Output Format block.

Advance the review mark so the cadence nudge resets and these reflections are not re-counted:

bun .agents/skills/algorithm-update/tools/mark-reviewed.ts

Close out: remind the user to commit the ALGORITHM.md change (it ships to all users on the next release) and to run pal cli install to activate it in their own runtime now.

Output format

Return:

The ranked candidate changes considered, with recurrence counts.

Which changes were applied to ALGORITHM.md and the exact edits.

Confirmation the review mark was advanced (new timestamp).

The commit + pal cli install reminder.

Do NOT use

Downstream / for a user's personal algorithm — this edits the shipped source. A personal algorithm overlay does not exist yet.

To edit ~/.pal/docs/ALGORITHM.md directly — always edit the repo source.