docs: show output path override example

[YOMXXX]

What problem are you trying to solve?

Issue #939 reports that project-level output path preferences can be missed when an agent follows the concrete Superpowers defaults for design specs and implementation plans.

The issue's working workaround is to put an explicit imperative instruction next to the Output Paths table. The maintainer direction on #939 was that the right fix is a README example showing how to override the default.

What does this PR change?

Adds a README-only Customizing Output Paths section that shows users how to configure alternate spec and plan locations in their project instruction file.

The example includes both the table and the direct MUST/NOT instruction from the working workaround, without changing any skill behavior.

Is this change appropriate for the core library?

Yes. This is general Superpowers user documentation for where core workflow artifacts are written. It is not project-specific, does not add a service or dependency, and does not change behavior-shaping skill content.

What alternatives did you consider?

I considered editing skills/brainstorming/SKILL.md and skills/writing-plans/SKILL.md, but that repeats the risky part of #1020: behavior-shaping skill edits without multi-session eval evidence.

I also considered adding a README regression test, but #1593 was closed partly because that test only grepped for strings the PR itself added. This PR intentionally keeps the change README-only, with no decorative test.

Does this PR contain multiple unrelated changes?

No. It changes one README section for one documentation gap: how to override the default output paths.

Existing PRs

• I have reviewed all open AND closed PRs for duplicates or prior art
• Related PRs: fix(brainstorming): honor CLAUDE.md Output Paths before defaults #1020, docs: show output path override example #1593

#1020 bundled skill edits with README documentation and did not follow the PR template. #1593 was closer, but it was part of a batch and added a tautological grep test. This PR is the narrower shape requested in the #1593 closure comment: README-only, no decorative regression test.

Environment tested

Harness (e.g. Claude Code, Cursor)	Harness version	Model	Model version/ID
Local shell/docs verification	zsh on macOS	n/a	n/a

New harness support (required if this PR adds a new harness)

Not applicable. This PR does not add support for a new harness.

Clean-session transcript for "Let's make a react todo list"

Not applicable: no new harness support is added by this PR.

Evaluation

• Initial problem source: Brainstorming skill ignores CLAUDE.md Output Paths, always writes to docs/superpowers/specs/ #939, including the reported workaround and the maintainer comment recommending a README example.
• This is documentation-only. I did not run agent behavior eval sessions because this PR does not change skill behavior.
• The README example documents the already-reported working form: an Output Paths table plus a direct MUST/NOT instruction next to it.

Verification run locally:

git diff --check
# no output

git diff --name-only | rg '^docs/superpowers/(specs|plans)/' || echo 'no generated spec/plan changes'
# no generated spec/plan changes

bash tests/opencode/run-tests.sh
# STATUS: PASSED, 2 passed, 0 failed

rg -n "Customizing Output Paths|Design specs MUST|Implementation plans MUST" README.md
# finds the new section and imperative instruction

Not run as a PR signal: node --test tests/pi/test-pi-extension.mjs. In this shell it cannot import .pi/extensions/superpowers.ts directly and failed the same way before the README edit.

Rigor

• If this is a skills change: I used superpowers:writing-skills and completed adversarial pressure testing (paste results below)
• This change was tested adversarially, not just on the happy path
• I did not modify carefully-tuned content (Red Flags table, rationalizations, "human partner" language) without extensive evals showing the change is an improvement

This is not a skills change. It deliberately avoids behavior-shaping skill edits and does not add a tautological docs test.

Human review

• A human has reviewed the COMPLETE proposed diff before submission

Fixes #939.