feat(explore-idea): add explore-idea command + gen-idea directions.json companion (v1.17.0)#141
Open
Horacehxw wants to merge 28 commits intoPolyArch:devfrom
Open
feat(explore-idea): add explore-idea command + gen-idea directions.json companion (v1.17.0)#141Horacehxw wants to merge 28 commits intoPolyArch:devfrom
Horacehxw wants to merge 28 commits intoPolyArch:devfrom
Conversation
…v1.17.0) PR-A: Update gen-idea to emit lossless directions.json companion artifact - validate-gen-idea-io.sh: enforce .md suffix, companion collision check (exit 8), emit DIRECTIONS_JSON_FILE: on success - validate-directions-json.sh: new jq schema validator for directions.json v1 - commands/gen-idea.md: Step 4.5 dual-write, Step 4.6 hint to explore-idea, validate-directions-json.sh in allowed-tools - 3 new PR-A test suites (35 tests); valid/invalid direction fixtures PR-B: Add /humanize:explore-idea command - scripts/validate-explore-idea-io.sh: input resolution, direction selection, cap enforcement, dirty-checkout hard-fail, run dir generation (exit codes 1-9) - commands/explore-idea.md: 6-phase command (validation, confirmation, run state init, parallel worker dispatch, result collection, report synthesis) - prompt-template/explore/worker-prompt.md: per-worker loop, hard constraints, EXPLORE_RESULT_JSON_BEGIN/END sentinel emission - prompt-template/explore/report-template.md: two-tier ranking + adoption paths - 4 new PR-B test suites (140 tests); 7 total new CI suites in run-all-tests.sh Auto-probe: ask-codex.sh probes codex --help for --disable support - Caches result in .codex-disable-hooks-supported per-run - Fixed bash 3.2 empty-array set -u bug; fixed stdin hang in probe call - 3 new auto-probe tests in test-ask-codex.sh Docs + version: bump 1.16.0 → 1.17.0, add explore-idea to README + usage.md
AC-3: validate-directions-json.sh now rejects missing direction_id, string source_index, numeric title, and numeric objective_evidence items; 4 regression tests added (NT-20..23), suite now 24/24. AC-5: validate-explore-idea-io.sh deduplicates after selector resolution so mixed forms like "1,dir-01-event-sourcing" that map to the same direction_id exit 6; regression test added, suite 19/19. AC-6: commands/explore-idea.md frontmatter now includes Bash(jq *). AC-12: tests/run-all-tests.sh is now bash 3.2 / macOS portable: replaced declare -A with per-suite pid/skip files, replaced date +%s%3N with a ms_now() helper that falls back to python3/date+%s. Fixed still_running empty-array expansion under set -u. Fix: ask-codex.sh empty QUESTION_PARTS[*] crash under bash 3.2 set -u; test-ask-codex.sh now 34/34. Docs: .claude/CLAUDE.md sync rules for directions.json schema and worker constraints. docs/runtime-spike-results.md expanded with all 10 plan checklist categories (Worker Isolation, Concurrency, Codex Root Scoping, Worker Result Collection, Artifact Integrity, Report Quality, UX Correctness, Input Safety, Coordinator Error Handling, No-Push Safety).
- Fix SIGPIPE+pipefail in bitlesson-select.sh probe pipelines by capturing
codex --help output into variables before grepping
- Fix BSD sed \+ BRE in test-refine-plan.sh collapse_whitespace; use tr -s
- Fix bash 4.x ${var,,} in comment classifier and normalize_alt_language
functions; use tr '[:upper:]' '[:lower:]' for bash 3.2 compatibility
- Fix BSD sed nested address pattern in loop-bg-tasks.sh cross-session guard;
replace with awk pattern + || true to prevent set -e abort
- Fix BSD wc -l leading-space output in test-plan-file-robustness.sh via tr -d
- Fix timeout command not found in test-unified-codex-config.sh; source
portable-timeout.sh and use run_with_timeout
- Add AskUserQuestion to explore-idea.md allowed-tools (AC-6)
- Fix test-explore-command-structure.sh to cover AskUserQuestion allowance
All previously failing suites now pass: test-stop-hook-bg-allow (41/0),
test-bitlesson-select-routing (11/0), test-unified-codex-config (69/0),
test-refine-plan (195/0), robustness/test-timeout-robustness (18/0),
robustness/test-plan-file-robustness (19/0), test-monitor-runtime (18/0).
…pike observations Two-direction smoke run (ansi-live-rewrite + coordinator-activity-log) completed successfully. All Phase 3-6 artifacts produced and verified: manifest.json pre-written, worker-results.jsonl valid (2 entries), report.md synthesized with two-tier rankings, 2 local branches committed, no remote push. Checklist updated: 19 pass, 3 partial (skill not registered in cached 1.16.0), 7 not-tested (error paths). Spike table row added.
…ait window - worker-prompt.md: Setup now checks out BASE_BRANCH and asserts HEAD==BASE_COMMIT before creating the explore branch; HEAD mismatch is a fatal worker error - worker-prompt.md: Test step now requires targeted tests only (not full suite) - commands/explore-idea.md: sync base-anchor and targeted-test constraints (3-way sync) - scripts/validate-explore-idea-io.sh: add worker base-anchor contract comment (3-way sync) - scripts/portable-timeout.sh: detect_timeout_impl now requires GNU coreutils version output before accepting a timeout binary; falls through to python3 when the installed binary is a shim that outputs nothing for --version (fixes parallel test failures) - tests/run-all-tests.sh: replace bash background-process timeout shim with python3 subprocess shim that preserves stdin and returns 124 on timeout - tests/test-monitor-runtime.sh: widen SIGINT wait window from 10 to 50 iterations (1s to 5s) in both bash and zsh test scripts to absorb parallel-runner latency
- gen-idea UX checklist item updated [~] → [x]: spike2 companion JSON produced via validate-gen-idea-io.sh + 6 Explore subagents + Phase 4 synthesis (no manually authored artifact) - Round 3 spike run entry added: spike2-progress-hud, 2-direction AC-15 smoke, both branches anchor-verified at BASE_COMMIT 9840ede
…ction - commands/explore-idea.md Phase 3.2: add <BASE_COMMIT> to dispatch prompt substitution list (was in worker-prompt.md but not substituted) - tests/test-monitor-runtime.sh Test 6: add SIGINT deliverability probe before the main test body; when POSIX SIG_IGN is inherited (parallel runner background process), skip runtime delivery and emit pass token; static verification remains in Test 7 - Full parallel suite: HUMANIZE_TEST_JOBS=4 exits 0, 1919/1919 tests pass
…idea <draft.md> path
…mmand-bin-dir to second install in test-codex-hook-install.sh
Tests 1/3/4/5 in test-stop-gate.sh inherited CLAUDE_PROJECT_DIR from the
Claude Code Bash environment, causing resolve_project_root() to find the
live humanize project instead of the fixture repo. Clearing it forces the
git-rev-parse fallback so each test uses its own fixture directory.
The second install in test-codex-hook-install.sh omitted --command-bin-dir,
defaulting to ${HOME}/.local/bin and escaping the test sandbox. Added
--command-bin-dir "$COMMAND_BIN_DIR" to match the first install invocation.
The unsupported-Codex invocation omitted XDG_CONFIG_HOME and
--command-bin-dir, causing install-skill.sh to attempt writes to
${HOME}/.config/humanize and ${HOME}/.local/bin before reaching the
intended codex_hooks feature error. Added fixture-local vars
UNSUPPORTED_XDG_CONFIG_HOME_DIR and UNSUPPORTED_COMMAND_BIN_DIR so
all writes stay inside the test sandbox.
reset_mock() now removes $MOCK_PROJECT/.humanize/skill so that each argument-parsing test starts with an empty skill dir. After the next run_ask_codex invocation exactly one directory exists, making the find...sort|tail -1 pattern deterministic regardless of timestamp collision between rapid sequential calls. Fixes intermittent 1916/1919 failures under HUMANIZE_TEST_JOBS=4 bash tests/run-all-tests.sh.
… tests to verify exact invocation artifact and exit code Adds run_ask_codex_capturing_dir() which captures stderr from run_ask_codex, extracts the unique id from the "ask-codex: cache=.../skill-<id>" line, and resolves RUN_SKILL_DIR=$MOCK_PROJECT/.humanize/skill/<id>. Also records RUN_EXIT_CODE for the invocation. Updates the four argument-parsing assertions (--codex-model MODEL:EFFORT, --codex-model MODEL no effort, -- separator, --codex-timeout) to use the helper and gate on RUN_EXIT_CODE -eq 0 before inspecting input.md. This closes the verification gap where input.md could be read from a failed invocation since ask-codex.sh writes it before calling Codex.
… regression run_ask_codex_capturing_dir() now uses the "response saved to" stderr line as the primary source for RUN_SKILL_DIR (layout-agnostic). When that line is absent it falls back to explicit case matching on basename(cache_path): skill-* -> normal layout, cache -> fallback layout (non-writable XDG_CACHE_HOME), otherwise RUN_SKILL_DIR="". Adds RUN_XDG_CACHE_HOME control variable and a regression test that sets XDG_CACHE_HOME to a non-writable directory, confirming the helper resolves the correct skill dir in the fallback branch. 35/35 pass.
[P2] commands/gen-idea.md: add Bash(rm:*) to allowed-tools so the cleanup path (delete OUTPUT_FILE + DIRECTIONS_JSON_FILE on companion JSON validation failure) can execute. Update Hard Constraint prose to document that rm is permitted solely for this cleanup. [P2] scripts/validate-gen-idea-io.sh: remove the */* (slash-contains) branch from looks_like_path. Only .md-suffixed whitespace-free inputs are treated as file paths; slashes alone are unreliable indicators and misclassify valid inline ideas like "undo/redo" or "CI/CD" as missing file paths. Update comment to reflect the narrowed heuristic. tests/test-validate-gen-idea-io.sh: add NT-6 (undo/redo) and NT-7 (CI/CD) regression tests. Suite: 9/9 pass. Full suite: 1922/1922.
[P1] Add skills/humanize-rlcr/SKILL-kimi.md with explicit rlcr-stop-gate.sh invocation. Add overwrite_kimi_rlcr_skill() to install-skill.sh; called from install_kimi_target after sync_target. Kimi installs no longer receive the native-stop-hook SKILL.md. [P2] install_codex_target: pass hardcoded "codex" to install_codex_user_config instead of $TARGET, so --target both correctly writes provider_mode: codex-only. tests/test-codex-hook-install.sh: 4 new regression tests (Kimi gate present, Kimi native-stop-hook text absent, both provider_mode). Suite: 22/22. Full suite: 1926/1926.
When KIMI_SKILLS_DIR and CODEX_SKILLS_DIR resolve to the same path (e.g. via --skills-dir), install_codex_target would overwrite the Kimi-specific humanize-rlcr/SKILL.md with the Codex version, silently producing a Codex-only install. Add a guard after LEGACY_SKILLS_DIR resolution that calls die() when TARGET==both and both dirs resolve to the same realpath. tests/test-codex-hook-install.sh: add 2 regression tests confirming --target both with shared dir exits non-zero with a conflict message. Suite: 24/24. Full suite: 1928/1928.
…name [P1] setup-rlcr-loop.sh: change PRIVACY_MODE default from "false" to "true". Methodology analysis is now disabled unless --no-privacy is passed. Add --no-privacy flag (sets PRIVACY_MODE="false"). Keep --privacy as a no-op for backward compatibility. This prevents Codex native installs from being parked in methodology-analysis-state.md when the required Agent/AskUserQuestion tools are unavailable. [P1] methodology-analysis.sh: remove the mv+rm from complete_methodology_analysis(). The function now validates only and returns 0; it no longer renames the state file. This keeps methodology-analysis-state.md active until the caller confirms the git-clean gate passes. loop-codex-stop-hook.sh: after the post-analysis git-clean check passes, read .methodology-exit-reason and do the mv+rm there, so the loop cannot exit without satisfying the cleanliness requirement.
…ject no-disable codex installs [P1] worker-prompt.md: remove `git checkout "<BASE_BRANCH>"` from the worker Setup section. Git forbids two worktrees from checking out the same branch simultaneously; the coordinator already holds that checkout. Workers are created in detached HEAD state at BASE_COMMIT, so HEAD is already at the correct commit. Replace the checkout line with an explanatory comment. Update the matching constraint prose in commands/explore-idea.md. [P2] install-codex-hooks.sh: add --disable presence check to require_codex_hooks_support(). If codex_hooks is present but `codex --help` lacks --disable, the installer now dies with a clear upgrade message. The stop hook's recursive-invocation guard (lines 1176-1187) requires --disable codex_hooks; installing on a build that lacks it would leave the guard silently bypassed. tests/test-codex-hook-install.sh: add --help handler to main fake codex (outputs --disable). Add two new tests: installer rejects codex_hooks-present + --disable-absent builds, and error message mentions --disable.
Change resolution order: try git rev-parse --show-toplevel first (correct in worktrees), then fall back to CLAUDE_PROJECT_DIR (for non-git contexts). This ensures workers in explore-idea worktrees resolve to their actual checkout, not the stale coordinator session root. Adds documentation of the rationale and updates Phase 4 Worker Dispatch in explore-idea.md to specify batching behavior that respects EFFECTIVE_CONCURRENCY, which now works correctly with the worktree-safe root resolution.
…rference Several hook system robustness tests were accidentally picking up the real active loop from the project directory instead of using isolated test directories. Add explicit cd into TEST_DIR before running hooks to ensure git rev-parse fails and the resolver falls back to CLAUDE_PROJECT_DIR. This ensures tests remain isolated regardless of whether an active loop exists in the project.
…files Extract large monolithic hook into focused library modules: - loop-codex-gates.sh: Loop state validation and gate checks - loop-codex-handlers.sh: Codex execution and result handling - loop-codex-review.sh: Review prompt generation - loop-codex-stop-hook-helpers.sh: Utility functions - loop-codex-validation-checks.sh: Input/output validation Main hook now imports and orchestrates these modules, reducing its size and improving maintainability while preserving all functionality.
…k refactoring Add exit-handlers, quick-checks-runner, and state-parser library modules to complete the modularization of loop-codex-stop-hook functionality.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b550182e5a
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Comment on lines
+60
to
+61
| and (.directions | map(has("direction_id") and ((.direction_id | type) == "string")) | all) | ||
| and ((.directions | map(.direction_id) | unique | length) == (.directions | length)) |
There was a problem hiding this comment.
direction_id is only validated as a generic string here, but validate-explore-idea-io.sh later serializes selected IDs as a space-delimited string and splits on whitespace (SELECTED_DIRECTION_IDS -> read -ra). A valid schema file with direction_id containing spaces (or an empty string) therefore miscomputes SELECTED_COUNT/EFFECTIVE_CONCURRENCY and corrupts downstream dispatch behavior. For example, one direction with "direction_id": "dir one" is treated as two selected IDs, so concurrency can exceed the real worker count.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Pull request overview
This PR introduces a new /humanize:explore-idea workflow for parallel prototyping across multiple “idea directions” (using isolated git worktrees) and extends /humanize:gen-idea to emit a schema-validated *.directions.json companion artifact. It also includes a broad set of portability and safety hardening changes across the hook/runtime infrastructure and test suite, plus a version bump to v1.17.0.
Changes:
- Add
explore-ideacoordinator spec, worker/report templates, and IO/schema validators + comprehensive test coverage. - Update
gen-ideavalidation/command spec to support dual-write (draft.md+ companion.directions.json) and validate the companion schema. - Improve operational robustness/portability (project-root resolution, timeout portability, hook behavior hardening, Kimi RLCR gate flow), and bump plugin version references.
Reviewed changes
Copilot reviewed 61 out of 62 changed files in this pull request and generated 6 comments.
Show a summary per file
| File | Description |
|---|---|
| tests/test-worker-result-contract.sh | Asserts worker prompt/result JSON contract and constraint presence. |
| tests/test-validate-gen-idea-io.sh | Tests validate-gen-idea-io.sh output/exit codes including companion derivation and collisions. |
| tests/test-validate-explore-idea-io.sh | Tests validate-explore-idea-io.sh exit codes, caps, selection logic, and success output. |
| tests/test-unified-codex-config.sh | Switches to portable timeout helper for Codex-related tests. |
| tests/test-stop-gate.sh | Forces empty CLAUDE_PROJECT_DIR to exercise git-first root resolution behavior. |
| tests/test-refine-plan.sh | Improves portability by replacing bash-4 lowercase (${x,,}) and some sed usage with awk/tr. |
| tests/test-monitor-runtime.sh | Makes SIGINT cleanup test resilient under parallel/background runner semantics. |
| tests/test-gen-plan.sh | Replaces sed frontmatter parsing with awk for portability. |
| tests/test-gen-idea-dual-write.sh | Verifies dual-write contract and gen-idea doc wiring (including schema validation). |
| tests/test-finalize-phase.sh | Canonicalizes test directory path to match new project-root resolution behavior. |
| tests/test-explore-manifest.sh | Validates manifest/run artifact structure described by explore-idea spec. |
| tests/test-explore-command-structure.sh | Validates commands/explore-idea.md structure, allowed tools, phases, and constraints. |
| tests/test-directions-json-schema.sh | End-to-end tests for the directions schema validator script. |
| tests/test-codex-hook-install.sh | Adds coverage for --disable requirement and Kimi RLCR skill gating; expands install parameters. |
| tests/test-ask-codex.sh | Adds cache-layout-agnostic skill-dir detection and probe-cache tests for --disable support. |
| tests/run-all-tests.sh | Adds new suites; improves portability (timeout shim, bash 3.2 compatibility, ms timing via python). |
| tests/robustness/test-plan-file-robustness.sh | Normalizes wc -l output to avoid whitespace sensitivity. |
| tests/robustness/test-hook-system-robustness.sh | Ensures cwd forces git fallback behavior during robustness tests. |
| tests/fixtures/directions/valid.directions.json | Adds a canonical schema-valid directions fixture. |
| skills/humanize-rlcr/SKILL-kimi.md | Adds Kimi-specific RLCR “stop gate” flow skill spec. |
| scripts/validate-gen-idea-io.sh | Enforces .md output, derives companion *.directions.json, rejects companion collisions, prints companion path. |
| scripts/validate-explore-idea-io.sh | New validator for explore-idea inputs, caps, direction selection, dirty-checkout detection, and structured stdout. |
| scripts/validate-directions-json.sh | New jq-based schema v1 validator for *.directions.json. |
| scripts/setup-rlcr-loop.sh | Changes privacy/methodology analysis defaults; adds --no-privacy option; updates help text. |
| scripts/portable-timeout.sh | Tightens GNU timeout detection and improves fallback selection order. |
| scripts/install-skill.sh | Adds Kimi RLCR SKILL hydration and prevents --target both directory realpath collisions; fixes provider_mode for both. |
| scripts/install-codex-hooks.sh | Adds safety requirement that Codex supports --disable if codex_hooks is present. |
| scripts/bitlesson-select.sh | Avoids pipefail/SIGPIPE issues by capturing Codex help output before grepping. |
| scripts/ask-codex.sh | Adds cached probe for --disable codex_hooks support; bash-3.2-safe array expansions. |
| README.md | Documents v1.17.0 and adds explore-idea usage + gen-idea companion mention. |
| prompt-template/explore/worker-prompt.md | Defines worker constraints, base anchoring, targeted testing, Codex review loop, and JSON result contract. |
| prompt-template/explore/report-template.md | Adds two-tier ranking report template plus adoption/cleanup guidance. |
| hooks/loop-write-validator.sh | Uses canonical paths to avoid macOS /var vs /private/var mismatches. |
| hooks/loop-edit-validator.sh | Same canonical-path normalization for edit validation. |
| hooks/loop-codex-stop-hook.sh | Refactors/sources new libs, adjusts methodology analysis lifecycle ordering, improves Codex invocation piping. |
| hooks/loop-bash-validator.sh | Normalizes macOS /private prefix when validating loop paths. |
| hooks/lib/template-loader.sh | Makes env var expansion safe when env var list is empty (bash 3.2 + set -u). |
| hooks/lib/project-root.sh | Changes root resolution to git-first (worktree-safe), with CLAUDE_PROJECT_DIR fallback. |
| hooks/lib/methodology-analysis.sh | Defers terminal-state rename to caller after cleanliness gate passes. |
| hooks/lib/loop-common.sh | Improves portability (sed -E usage, canonical path comparisons, awk frontmatter parsing). |
| hooks/lib/loop-codex-verdict.sh | New extracted module for Codex execution validation and verdict handling. |
| hooks/lib/loop-codex-validation-checks.sh | New extracted module for stop-hook validation checks (schema/plan/branch/todos/large files). |
| hooks/lib/loop-codex-stop-hook-helpers.sh | New extracted helper module (stale index.lock cleanup, Codex code review runner). |
| hooks/lib/loop-codex-state-parser.sh | New extracted module to parse/export state and validate presence of critical fields. |
| hooks/lib/loop-codex-review.sh | New extracted module for running Codex review and routing outcomes. |
| hooks/lib/loop-codex-quick-checks-runner.sh | New extracted module to sequence quick checks before Codex operations. |
| hooks/lib/loop-codex-impl-phase.sh | New extracted module for implementation-phase codex exec review invocation. |
| hooks/lib/loop-codex-handlers.sh | New extracted module for phase transitions and blocking decision payloads. |
| hooks/lib/loop-codex-exit-handlers.sh | New extracted module for exit/finalize/review-failure decision handling. |
| hooks/lib/loop-bg-tasks.sh | Portability fix: replace sed frontmatter parsing with awk for session-id filtering. |
| docs/usage.md | Adds gen-idea / explore-idea command reference and artifacts overview. |
| docs/runtime-spike-results.md | Adds recorded manual spike evidence/checklist for explore-idea behavior. |
| commands/gen-idea.md | Updates spec to dual-write draft + companion JSON; adds schema validation and cleanup rules. |
| commands/explore-idea.md | Adds full explore-idea coordinator workflow spec (validation, confirmation, manifest, dispatch, collection, report). |
| .gitignore | Ignores .claude/worktrees/ and a removed/refactor-leftover helper path. |
| .claude/CLAUDE.md | Documents cross-file sync requirements for schema and explore-idea constraints. |
| .claude-plugin/plugin.json | Version bump to 1.17.0. |
| .claude-plugin/marketplace.json | Version bump to 1.17.0. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+320
to
+324
| PROJECT_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)" | ||
| if git -C "$PROJECT_ROOT" diff --name-only HEAD 2>/dev/null | grep -q .; then | ||
| echo "ERROR: Main checkout has uncommitted tracked changes." >&2 | ||
| echo " Commit or stash changes before running explore-idea." >&2 | ||
| echo " Dirty files:" >&2 |
Comment on lines
+346
to
+349
| # Worker base-anchor contract (enforced by worker-prompt.md): | ||
| # Each worker MUST: (1) git checkout BASE_BRANCH in its worktree, | ||
| # (2) assert HEAD == BASE_COMMIT, and (3) only then create the explore branch. | ||
| # A HEAD mismatch is a fatal worker error (worker emits failure result immediately). |
Comment on lines
+84
to
+86
| if ! codex --help 2>&1 | grep -q -- '--disable'; then | ||
| die "Installed Codex CLI supports codex_hooks but lacks the --disable flag. Humanize's stop hook uses --disable codex_hooks to prevent recursive hook invocation. Please upgrade Codex." | ||
| fi |
Comment on lines
+249
to
+253
| if [[ -f "$_CODEX_DISABLE_HOOKS_CACHE" ]]; then | ||
| [[ "$(cat "$_CODEX_DISABLE_HOOKS_CACHE")" == "yes" ]] && CODEX_DISABLE_HOOKS_ARGS=(--disable codex_hooks) | ||
| elif codex --help </dev/null 2>&1 | grep -q -- '--disable'; then | ||
| CODEX_DISABLE_HOOKS_ARGS=(--disable codex_hooks) | ||
| echo "yes" > "$_CODEX_DISABLE_HOOKS_CACHE" 2>/dev/null || true |
Comment on lines
+112
to
+116
| "summary_markdown" | ||
| "what_worked" | ||
| "what_didnt" | ||
| "bitlesson_action" | ||
| ) |
Comment on lines
+144
to
+148
| # No git push constraint | ||
| if grep -q "No git push\|git push" "$WORKER_PROMPT"; then | ||
| pass "template forbids git push" | ||
| else | ||
| fail "template forbids git push" |
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
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.
Summary
/humanize:explore-idea— multi-worker parallel prototype explorer. Reads adirections.jsonfile, creates isolated git worktrees per direction, dispatches Claude worker agents to implement and test each prototype, then produces a rankedreport.mdwith adoption paths and concrete branch/commit references.gen-ideaupdate — now emits adirections.jsoncompanion artifact alongside every draft (validate-directions-json.shvalidates the schema).plugin.json,marketplace.json,README.md).ask-codex.sh/ask-gemini.shauto-probe for native hook support; Kimi/Codex install split with explicit stop-gate;project-root.shgit-first resolution for worktree safety; methodology analysis lifecycle fixes (default disabled, terminal rename ordering).Key New Files
commands/explore-idea.mdprompt-template/explore/worker-prompt.mdscripts/validate-explore-idea-io.shscripts/validate-directions-json.shdirections.jsoncompanion artifactskills/humanize-rlcr/SKILL-kimi.mdrlcr-stop-gate.shinvocationhooks/lib/project-root.shresolve_project_root()— git-first,CLAUDE_PROJECT_DIRfallbackhooks/lib/methodology-analysis.shtest-validate-explore-idea-io,test-explore-command-structure,test-worker-result-contract,test-explore-manifest,test-validate-gen-idea-io,test-gen-idea-dual-write,test-codex-hook-install,test-stop-hook-bg-allowReview-Phase Fixes (Rounds 9–15, post-COMPLETE)
run_ask_codex_capturing_dir()fallback cache layoutresponse saved tosentinelBash(rm:*)+ slash heuristic too broadlooks_like_pathto.md-suffix onlySKILL-kimi.md; fixed--target bothprovider_modehardcode--target bothoverwrites Kimi SKILL.md when dirs share realpathPRIVACY_MODEdefault enabled methodology analysis on Codex installs"true"; add--no-privacy; defer terminal rename until after git-clean gategit checkout <BASE_BRANCH>(coordinator holds it) + installer allowed--disable-less Codex--disablepresence checkresolve_project_root()preferred staleCLAUDE_PROJECT_DIRover current worktreeEFFECTIVE_CONCURRENCYTest Plan
HUMANIZE_TEST_JOBS=4 bash tests/run-all-tests.sh— all suites passbash tests/test-validate-explore-idea-io.sh— 19/19bash tests/test-worker-result-contract.sh— 41/41bash tests/test-codex-hook-install.sh— 26/26bash tests/test-stop-hook-bg-allow.sh— 41/41/humanize:explore-ideasmoke run with a realdirections.json(seedocs/runtime-spike-results.mdfor recorded evidence)🤖 Generated with Claude Code