agentleFS
nshopiknshopik/dream-team/skills/friction/SKILL.md

friction

Skill2 starsChanged 19 days agoMITView on GitHub

Scans past build-issue-loop workflow runs for agent friction (skill or tool hunting under ~/.claude, denied or erroring tool calls, repeated failing commands, forbidden actions, self-reported friction, token and tool-call outliers) and reports the patterns seen in two or more runs, each with the deterministic check that would stop it or the prompt line it points to. Use when the user runs /dream-team:friction or asks what keeps going wrong across build-issue runs.

  • Commits and pushes

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the evidence before blaming a prompt or naming a check.

    line 46
  2. Distinctive:

    No check and no prompt line behind it → list it as ordinary and offer nothing for it.

    line 52
  3. Very specific:

    No check, a prompt line → <target> is <file>:<line>.

    line 58

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

friction

1. Scan

  • Run python3 <this skill's base directory>/friction.py.
  • User asks for run times → add --time; print its table and stop.
  • The script scans every workflow run not yet in the cache and appends its hits there.
  • It prints each pattern + role seen in 2 or more runs; one-off hits stay in the cache only.
  • Output is only the scanned … new runs line → report no repeated friction and stop.

2. Trace each pattern

Pattern names the script prints:

  • claude-dir: Read/Glob/Grep, or a Bash find/ls/cat, under ~/.claude.
  • tool-error: a tool call denied or erroring, other than a plain non-zero exit.
  • repeat-failure: the same failing Bash command run twice or more by one agent.
  • forbidden: ssh, git push, a gh/glab write, or an edit, a Bash tree edit (sed/perl -i, git checkout/restore/stash, tee or a redirect to a path outside /tmp/ and /dev/) or a commit by a read-only role.
  • self-report: an entry from the agent's friction output field.
  • tokens-outlier, tools-outlier: over 2x the role's median over earlier runs.

For each printed pattern:

  • Find the prompt that role runs in <base directory>/../../workflows/build-issue-loop.js: impl → implPrompt, gate → verifyPrompt, build-fix → buildFixPrompt, simplify → simplifyPrompt, simplify-fix → simplifyFixPrompt, review:domain → domainPrompt, review:external → externalPrompt, other review: → lensPrompt, re-review: → reReviewPrompt, fixdiff: → fixDiffPrompt, fix → fixPrompt or nitFixPrompt.
  • Name the prompt or skill line the evidence points to: the instruction the agent could not follow, or the tool it lacked.
  • Read the evidence before blaming a prompt or naming a check; a hook bounce working as designed or a long run on a hard issue has neither behind it.
  • A permission rule, hook or gate command could stop the pattern → name that check and what it stops.
  • The pattern needs judgement → name no check.
  • The only check is a new hook and no written rule forbids what the agent did → name no check.
  • No check and no prompt line behind it → list it as ordinary and offer nothing for it.

3. Report

  • One bullet per pattern: - **<pattern> | <role>:** <n> runs; <evidence>; <target>.
  • A named check → <target> is check: <check>.
  • No check, a prompt line → <target> is <file>:<line>.
  • Ordinary → <target> is ordinary.
  • Offer to file one dream-team issue per pattern with a check or a prompt line, for its <target>, written with scope-issue.
  • File nothing without the user's yes.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.