refactor: strip social proof, self-selling, and recap detritus from 12 skills (eval-gated)#1934
Draft
obra wants to merge 12 commits into
Draft
refactor: strip social proof, self-selling, and recap detritus from 12 skills (eval-gated)#1934obra wants to merge 12 commits into
obra wants to merge 12 commits into
Conversation
Real-World Impact restated the Real Example from Session as statistics; Key Benefits and the time-saved line sold the skill to a reader already executing it. Instructions unchanged.
Real-World Impact was statistics; the Overview opener restated the core principle as motivation. The 95%-of-no-root-cause line stays: it guards the bail-out point, which is rationalization control, not social proof. Supporting Techniques/Related skills untouched (PR #1932 owns that).
…ompletion Why This Matters (failure-memory testimonials), the dishonesty reframing in the Overview, and The Bottom Line recap all restate stakes the Iron Law, gate function, and rationalization table already enforce. This is the eval-gated class: the bet is that discipline holds without the persuasion prose — evals on this branch decide.
The tell-your-partner directive and the prefer-SDD instruction stay; the significantly-higher-quality sentence restated them as a claim. Integration section untouched (PR #1932 owns it).
…pment Five blocks of benefits and cost/benefit selling aimed at a reader who has already invoked the skill; the vs-Executing-Plans comparison also duplicates the one under When to Use. Integration section untouched (PR #1932 owns it).
…view Integration with Workflows restated the When to Request Review triggers grouped by caller (each-task/before-merge/when-stuck all appear at point of use), and the intro's mechanism-rationale sentence sold a rule the preceding sentence already states.
…nalization table Common Mistakes and Red Flags restated Steps 0-3 wholesale; both fold into one Common Rationalizations table (house Excuse/Reality form) whose five rows carry the tempting-thought version of each rule, including the #1-mistake emphasis on bypassing native tools. Quick Reference stays as the compact decision aid.
Five of six principles restated the Checklist and Process sections verbatim-in-spirit. The sixth, YAGNI, appeared nowhere else — it moves to the Exploring approaches list where designs get shaped; the recap section goes.
All four lines restate the Overview (DRY/YAGNI/TDD/frequent commits), Task Structure (exact paths, commands with expected output), and No Placeholders (complete code in every step).
Restates the Iron Law, the RED-GREEN-REFACTOR mapping, and the TDD-for-docs framing, all stated in full earlier in the file.
Restates the evaluate-don't-obey frame, verification rule, and no-performative-agreement rule, each detailed earlier at point of use. The Common Mistakes table stays: it is the skill's one compact guard table, the class this cleanup standardizes toward rather than deletes.
…evelopment The section's five prose rebuttals each map to a Common Rationalizations row (test-after, manual-tested, sunk-cost, dogmatic, spirit-not-ritual), and every excuse phrasing also appears in the Red Flags list. This is the highest-stakes cut on the branch: TDD is the most pressure-tested discipline skill, and the bet that the table alone holds under pressure is exactly what the eval pass must decide.
5 tasks
This was referenced Jul 6, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Who is submitting this PR? (required)
What problem are you trying to solve?
Skills carry content that burns context tokens on every load without
shaping behavior: social-proof sections ("Real-World Impact" with
15-30-minute/95% statistics), self-selling sections ("Advantages",
"Key Benefits") aimed at a reader who has already invoked the skill, and
end-of-file recaps ("Remember", "The Bottom Line", "Key Principles") that
restate rules already at point of use. A 14-agent audit (one auditor per
skill, shared brief distinguishing detritus from pressure-tested behavior
control) surveyed every skill; every finding on this branch was then
verified by hand against the claimed point-of-use duplicate before
cutting — the audit's numbers were not trusted blind (one was inflated
2.6x).
What does this PR change?
Twelve commits, one skill each, net −222/+11 lines:
(Real-World Impact, Key Benefits), systematic-debugging (Real-World
Impact, Overview throat-clearing), subagent-driven-development
(Advantages), executing-plans (quality claim), requesting-code-review
(workflow index + mechanism-rationale sentence).
use: brainstorming (Key Principles — with YAGNI, a sole carrier the
audit missed, moved into Exploring approaches first), writing-plans
(Remember), writing-skills and receiving-code-review (The Bottom Line).
Common Rationalizations table (house Excuse/Reality form), preserving
the Add problem-solving skills from amplifier patterns #1-mistake emphasis on bypassing native worktree tools.
the dishonesty reframing, and The Bottom Line;
test-driven-development loses Why Order Matters (five prose rebuttals,
each mapped to an existing Common Rationalizations row and Red Flags
entry before deletion).
Judgment calls that diverge from the raw audit, for reviewers to check:
systematic-debugging's "95% of no-root-cause cases" line STAYS (it guards
the bail-out point — rationalization control, not social proof);
receiving-code-review's Common Mistakes table STAYS (one compact guard
table per skill is the pattern this cleanup standardizes toward, not
zero).
Is this change appropriate for the core library?
Yes — deletions and one table conversion in core skills; no new
dependencies or integrations.
What alternatives did you consider?
per-skill commits is the unit the eval pipeline can bisect.
it leaves the actual question — does persuasion prose in discipline
skills do anything? — unanswered. Landing them on the eval branch is
how the question gets answered; they are trivially revertable
per-commit.
found a haiku auditor had inflated a section's size ~2.6x; every cut
was re-verified against the file by hand.
Does this PR contain multiple unrelated changes?
Twelve skills, one theme: removing content that does not change agent
behavior, found by one audit, structured for one eval run. The commits
are separable on purpose — reverting any single commit restores that
skill exactly.
Existing PRs
deliberately leaves those sections alone in SDD, executing-plans, and
systematic-debugging to avoid overlap; both branches touch the same
files, whichever lands second sees trivial or no conflicts), refactor: modernize finishing-a-development-branch (discard demotion, worktree-path fix, rationalization table) #1933
(finishing-a-development-branch modernization — that skill is untouched
here; its audit ran against the rewritten version and came back clean),
feat: agentic-end-to-end-testing skill + spec-derived scenario cards #1931 (agentic-e2e — touches SDD's SKILL.md in regions this branch does
not).
Environment tested
New harness support (required if this PR adds a new harness)
N/A — no new harness.
Evaluation
like social proof 'real world impact'... we should audit all the skills
for that... fan out haiku subagents one per skill."
with an explicit protected-content list (Iron Laws, rationalization
tables, thought-tables, menus, digraphs) and the test "if this text
vanished, would an agent act differently?"
the point-of-use duplicate the auditor claimed (grep/read), which
caught one inflated size claim, one misclassified guard line, and one
sole-carrier (YAGNI) that required a move instead of a delete.
entire purpose and the pre-merge gate for @arittr. The discipline-skill
cuts (TDD, verification-before-completion) should get the heaviest
pressure scenarios.
Rigor
superpowers:writing-skillsandcompleted adversarial pressure testing (paste results below)
rationalizations, "human partner" language) without extensive evals
showing the change is an improvement
The two unchecked boxes are honest: this branch modifies pressure-tested
discipline skills ahead of the evals that would justify it, by design —
the maintainer chose to structure the question as one eval-able branch.
No rationalization-table rows or "human partner" language were changed;
the cuts remove prose adjacent to those tables. Nothing here merges
without the eval pass.
Human review
Draft PR: opened for @obra's full-diff review and @arittr's eval pass; box
intentionally unchecked until that review happens.