Skip to content

feat(refs): base-instructions — communication anti-patterns, scope management (Level 0→3)#632

Open
notque wants to merge 1 commit into
mainfrom
enrich/refs-2026-05-10-0407
Open

feat(refs): base-instructions — communication anti-patterns, scope management (Level 0→3)#632
notque wants to merge 1 commit into
mainfrom
enrich/refs-2026-05-10-0407

Conversation

@notque
Copy link
Copy Markdown
Owner

@notque notque commented May 10, 2026

Reference Enrichment: base-instructions

Summary

Enriches agents/base-instructions from Level 0 (no references) to Level 3 (deep).

Targets Processed

Agent Before After New Files
base-instructions Level 0 Level 3 2

New Reference Files

  • agents/base-instructions/references/communication-anti-patterns.md (232 lines)
    Concrete prohibited phrase table, grep-detectable anti-pattern catalog (self-congratulation, narration openers, hedged outcomes, trailing validation requests, capability framing), error-fix mappings, and correct prose pattern examples.

  • agents/base-instructions/references/scope-management.md (263 lines)
    Over-engineering signal table, grep/rg detection commands for premature abstraction, impossible nil guards, compat shims, feature flags, and unrequested test additions. Includes a scope verification checklist using git diff.

Loading Table

Added to agents/base-instructions.md — maps task signals to reference files:

  • Verbose/hedged/self-congratulatory output → communication-anti-patterns.md
  • Scope wider than requested; abstractions or shims added → scope-management.md
  • Diff contains single-use helpers, nil guards, unrequested test files → scope-management.md

Audit Result

[agent] base-instructions  Level 3 (Deep)
  communication-anti-patterns.md   232 lines  concrete=90  generic=0  code=Y  cmds=Y
  scope-management.md              263 lines  concrete=99  generic=0  code=Y  cmds=Y

Validation Gate

[KEEP] base-instructions: Loading table correctly routes all 3 test signal types; both files contain grep-detectable patterns with bad/good code pairs — not generic advice.


Generated by reference-enrichment headless process (ADR-173) — Run ID: 2026-05-10-0407

…nagement (Level 0→3)

Adds two Level-3 reference files to agents/base-instructions:

- communication-anti-patterns.md: grep-detectable prohibited phrases with
  bad/good prose pairs (self-congratulation, hedging, narration, trailing
  validation requests). 232 lines, 90 concrete hits, zero generic advice.

- scope-management.md: over-engineering anti-patterns with rg/grep detection
  commands — premature abstraction, impossible nil guards, compat shims,
  feature flags, unrequested tests. 263 lines, 99 concrete hits.

Updates agents/base-instructions.md with a loading table mapping task
signals (verbose output, scope creep, diff review) to the two new files.

Audit result: Level 0 → Level 3 (Deep).
@notque notque enabled auto-merge (squash) May 10, 2026 04:14
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.

1 participant