Skip to content

refactor(pstack): distill skill prompts to cut tokens - #538

Merged
cursor[bot] merged 1 commit into
mainfrom
lauren/pstack-distill-port-3889
Oct 10, 2026
Merged

cursor[bot] merged 1 commit into
mainfrom
lauren/pstack-distill-port-3889

Conversation

@poteto

@poteto poteto commented Oct 10, 2026

Copy link
Copy Markdown
Collaborator

Summary

Tightens the prose in 35 pstack skill, reference, and playbook files. Every rule stays. Only the text that carries it gets shorter.

Across the touched Markdown files, size drops from 169,348 to 153,004 bytes (-9.7%) and from 26,177 to 23,829 words (-9.0%, counted with wc -w).

Changes

  • Principle skills: shorter descriptions and bodies for ten principle-* skills.
  • poteto-mode: the Principles index states the skill naming rule once instead of repeating each skill name. The PR-opening and trace-forensics playbooks are tighter.
  • how, why, reflect, interrogate, arena, figure-it-out, benchmark-checklist, technical-writing, unslop: steps and reference prompts are compressed.
  • architect: Phase A asked for the "traced model" that how prescribes, but how produces an architectural explanation. It now asks for that. The rest of the skill is compressed.
  • Bumps pstack to 0.15.16.

Model defaults (Opus at xhigh, Grok at xhigh-fast), panel makeup, /poteto-help, and /setup-pstack do not change.

Test plan

  • node scripts/validate-plugins.mjs passes.
  • Every entry in the poteto-mode Principles index resolves to an existing principle-* skill folder under the new naming rule (24 of 24).
  • Model slugs in edited lines match the current defaults.
Open in Web Open in Cursor 

Tighten the prose in 35 skill, reference, and playbook files without
dropping any rule. Total size drops by about 10% in bytes and 9% in
words.

/architect now asks for the architectural explanation that /how
produces, not a traced model that /how never makes.

Bump pstack to 0.15.16.
@cursor
cursor Bot merged commit 00b14f1 into main Oct 10, 2026
2 checks passed
hugo-hsi-dev added a commit to hugo-hsi-dev/ostack that referenced this pull request Oct 12, 2026
…hes (#31)

* Re-port the files whose text predates pstack 0.15.20

The last two hand syncs left upstream text from before pstack 0.15.20 in
21 files. These are the 15 files cursor/plugins#538 distilled, onyo-mode (cursor/plugins#538, cursor/plugins#575),
opening-a-pr (cursor/plugins#538), babysit (cursor/plugins#571), worktree-cleanup (cursor/plugins#571, cursor/plugins#574),
create-verification-skill (cursor/plugins#573), and README's onyo-agent paragraph (cursor/plugins#575).

Each file now starts from the pinned upstream text, with ostack's harness
port, defaults, and Projects edits applied again on top.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LLt7YKP4TztKihc9YUneAj

* Generate pstack-origin files from upstream plus port patches

sync/sync.py rebuilds each of ostack's 167 pstack-origin files from upstream
pstack at the commit in sync/PIN. It applies the harness map, removes
disable-model-invocation from skills outside KEEP_FLAG, and applies the
file's patch in sync/patches. Every patch hunk is marked with its audit code
([A] harness port, [B] user default, [C] Projects workflow, [D] ostack's own
change), so later work can find and remove one kind of change by grepping.

check fails when a file differs from that result. refresh rewrites a patch
after a hand edit and keeps the markers of hunks the edit didn't touch. bump
merges ostack's changes onto a new upstream commit with git merge-file and
lists the conflicts. sync/README.md explains each step.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LLt7YKP4TztKihc9YUneAj

* Run the sync check and the onyo-mode hook test in CI

The ci workflow runs on pushes to main and on pull requests. Its sync job
fails when a generated file no longer matches upstream plus its patch. Its
hooks job runs hooks/onyo-mode.test.sh.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LLt7YKP4TztKihc9YUneAj

---------

Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants