fix(pattern-mapper): prevent redundant file reads and add early-stop rule (#2312) (#2327)

* feat: add /gsd-spec-phase — Socratic spec refinement with ambiguity scoring (#2213)

Introduces `/gsd-spec-phase <phase>` as an optional pre-step before discuss-phase.
Clarifies WHAT a phase delivers (requirements, boundaries, acceptance criteria) with
quantitative ambiguity scoring before discuss-phase handles HOW to implement.

- `commands/gsd/spec-phase.md` — slash command routing to workflow
- `get-shit-done/workflows/spec-phase.md` — full Socratic interview loop (up to 6
  rounds, 5 rotating perspectives: Researcher, Simplifier, Boundary Keeper, Failure
  Analyst, Seed Closer) with weighted 4-dimension ambiguity gate (≤ 0.20 to write SPEC.md)
- `get-shit-done/templates/spec.md` — SPEC.md template with falsifiable requirements
  (Current/Target/Acceptance per requirement), Boundaries, Acceptance Criteria,
  Ambiguity Report, and Interview Log; includes two full worked examples
- `get-shit-done/workflows/discuss-phase.md` — new `check_spec` step detects
  `{padded_phase}-SPEC.md` at startup; displays "Found SPEC.md — N requirements
  locked. Focusing on implementation decisions."; `analyze_phase` respects `spec_loaded`
  flag to skip "what/why" gray areas; `write_context` emits `<spec_lock>` section
  with boundary summary and canonical ref to SPEC.md
- `docs/ARCHITECTURE.md` — update command/workflow counts (74→75, 71→72)

Closes #2213

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(pattern-mapper): prevent redundant file reads and add early-stop rule (#2312)

Adds three explicit constraints to the agent prompt:
1. Read each analog file EXACTLY ONCE (no re-reads from context)
2. For files > 2,000 lines, use Grep + Read with offset/limit instead of full load
3. Stop analog search after 3–5 strong matches

Also adds <critical_rules> block to surface these constraints at high salience.
Adds regression tests READS-01, READS-02, READS-03.

Closes #2312

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(pattern-mapper): clarify re-read rule allows non-overlapping targeted reads (CR feedback)

"Read each file EXACTLY ONCE" conflicted with the large-file targeted-read
strategy. Rewrites both the Step 4 guidance and the <critical_rules> block to
make the rule precise: re-reading the same range is forbidden; multiple
non-overlapping targeted reads for large files are permitted.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Tom Boucher
2026-04-16 17:15:29 -04:00
committed by GitHub
parent 0da696eb6c
commit 2acb38c918
2 changed files with 48 additions and 1 deletions

View File

@@ -118,6 +118,12 @@ Grep("router\.(get|post|put|delete)", type: "ts")
## Step 4: Extract Patterns from Analogs
**Never re-read the same range.** For small files (≤ 2,000 lines), one `Read` call is enough — extract everything in that pass. For large files, multiple non-overlapping targeted reads are fine; what is forbidden is re-reading a range already in context.
**Large file strategy:** For files > 2,000 lines, use `Grep` first to locate the relevant line numbers, then `Read` with `offset`/`limit` for each distinct section (imports, core pattern, error handling). Use non-overlapping ranges. Do not load the whole file.
**Early stopping:** Stop analog search once you have 35 strong matches. There is no benefit to finding a 10th analog.
For each analog file, Read it and extract:
| Pattern Category | What to Extract |
@@ -297,6 +303,16 @@ Pattern mapping complete. Planner can now reference analog patterns in PLAN.md f
</structured_returns>
<critical_rules>
- **No re-reads:** Never re-read a range already in context. Small files: one Read call, extract everything. Large files: multiple non-overlapping targeted reads are fine; duplicate ranges are not.
- **Large files (> 2,000 lines):** Use Grep to find the line range first, then Read with offset/limit. Never load the whole file when a targeted section suffices.
- **Stop at 35 analogs:** Once you have enough strong matches, write PATTERNS.md. Broader search produces diminishing returns and wastes tokens.
- **No source edits:** PATTERNS.md is the only file you write. All other file access is read-only.
- **No heredoc writes:** Always use the Write tool, never `Bash(cat << 'EOF')`.
</critical_rules>
<success_criteria>
Pattern mapping is complete when:

View File

@@ -1,11 +1,12 @@
/**
* Tests for Pattern Mapper feature (#1861)
* Tests for Pattern Mapper feature (#1861, #2312)
*
* Covers:
* - Config key workflow.pattern_mapper in VALID_CONFIG_KEYS
* - Default value is true
* - Config round-trip (set/get)
* - init plan-phase output includes patterns_path (null when missing, path when present)
* - Agent prompt contains no-re-read and early-stop constraints (#2312)
*/
const { describe, test, beforeEach, afterEach } = require('node:test');
@@ -109,3 +110,33 @@ describe('init plan-phase patterns_path', () => {
assert.ok(data.patterns_path.includes('01-foundation'), `Expected path to include phase dir, got: ${data.patterns_path}`);
});
});
describe('gsd-pattern-mapper agent prompt efficiency constraints (#2312)', () => {
const agentPath = path.join(__dirname, '..', 'agents', 'gsd-pattern-mapper.md');
let agentContent;
beforeEach(() => {
agentContent = fs.readFileSync(agentPath, 'utf-8');
});
test('READS-01: prompt contains no-re-read constraint', () => {
assert.ok(
/read each.*file.*once/i.test(agentContent) || /never re-read/i.test(agentContent),
'Agent prompt must instruct the model to read each analog file only once'
);
});
test('READS-02: prompt contains early-stop instruction', () => {
assert.ok(
/stop.*analog|3.?5.*analog|early.stop/i.test(agentContent),
'Agent prompt must instruct the model to stop after finding 3-5 analogs'
);
});
test('READS-03: prompt contains large-file strategy', () => {
assert.ok(
/2[,.]?000.*line|offset.*limit|large file/i.test(agentContent),
'Agent prompt must include guidance for reading large files with offset/limit'
);
});
});