feat: redesign empty-queue behavior — stay in repo, audit, present, stop#9
Merged
Conversation
When the task queue is empty, agents now: 1. Clean up the current repo (merge PRs, prune worktrees) 2. Run a project audit (lint, tests, security, dead code, etc.) 3. Present findings to the user as candidate tasks 4. Stop and wait for direction Removed the `find ~/apps` cross-repo search that caused agents to drift into unrelated codebases. Updated all 6 command variants via generate-commands and the README step list. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
6 tasks
fyodoriv
added a commit
that referenced
this pull request
May 4, 2026
…ess/Pivot/Measurement/Anchor + Estimate/Verification/Risk) (#78) Promote 8 fields Minsky has been treating as "custom" to first-class metadata and add a dedicated **Rule-#9 pre-registration block** section to spec.md so non-trivial changes can declare their predicted observable *before* the code is written — preventing post-hoc fishing for flattering metrics (Munafò et al. 2017) and pre-registering the give-up criterion (Ries 2011, pivot-or-persevere). Updates README, the canonical `commands/next-task.md`, and regenerates the 6 variant SKILL files. Co-authored-by: Your Name <your@email.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
8 tasks
fyodoriv
added a commit
that referenced
this pull request
May 21, 2026
Replaces the "Plan (complex tasks only)" section in `commands/next-task.md` with a 5-step "Plan and validate" workflow: 1. Check for an existing validated plan at `docs/plans/<task-id>.md` 2. Write the plan from `examples/plan-template.md` 3. Validate with a reviewer subagent (`reviewer` profile, fallback chain `qa-engineer` → `researcher`); the subagent appends a `## Reviewer verdict` block — only `approved` lets work proceed; `needs-revision` triggers up to 3 revision cycles; `reject` halts and reports 4. Pre-register the rule-#9 metric in the task block 5. Implement Trivial tasks (single file, <30 minutes, obvious fix) keep skipping the plan step. The legacy `**Plan**:` inline-checklist field in TASKS.md is superseded for the canonical workflow; the linter still accepts it for backwards compat. Why: today the same agent both writes and executes the plan. No second opinion. A separate plan file + reviewer-subagent validation makes deliberation external, diffable, and re-runnable. See the plan in `docs/plans/next-task-plan-first-workflow.md` for the full motivation and risk analysis. This commit also: - Updates `commands/lint-tasks.md` with a prose-only convention paragraph (no linter binary change; hard CI gate deferred to follow-up task `lint-tasks-md-plan-required-gate`). - Updates `AGENTS.md` "Task Queue Policy" section to point at the new workflow instead of the legacy `**Plan**:` checklist. - Updates `README.md` "What it does" step 12 to mention the new flow. - Regenerates all 12 agent variants via `npx tasks generate-commands` so `commands-drift` CI stays green. Verification: `npm run build`, `npm test` (238 tests across 6 packages, all green), `npm run lint`, `npx -y @tasks-md/lint TASKS.md` all green.
fyodoriv
added a commit
that referenced
this pull request
May 21, 2026
Replaces the "Plan (complex tasks only)" section in `commands/next-task.md` with a 5-step "Plan and validate" workflow: 1. Check for an existing validated plan at `docs/plans/<task-id>.md` 2. Write the plan from `examples/plan-template.md` 3. Validate with a reviewer subagent (`reviewer` profile, fallback chain `qa-engineer` → `researcher`); the subagent appends a `## Reviewer verdict` block — only `approved` lets work proceed; `needs-revision` triggers up to 3 revision cycles; `reject` halts and reports 4. Pre-register the rule-#9 metric in the task block 5. Implement Trivial tasks (single file, <30 minutes, obvious fix) keep skipping the plan step. The legacy `**Plan**:` inline-checklist field in TASKS.md is superseded for the canonical workflow; the linter still accepts it for backwards compat. Why: today the same agent both writes and executes the plan. No second opinion. A separate plan file + reviewer-subagent validation makes deliberation external, diffable, and re-runnable. See the plan in `docs/plans/next-task-plan-first-workflow.md` for the full motivation and risk analysis. This commit also: - Updates `commands/lint-tasks.md` with a prose-only convention paragraph (no linter binary change; hard CI gate deferred to follow-up task `lint-tasks-md-plan-required-gate`). - Updates `AGENTS.md` "Task Queue Policy" section to point at the new workflow instead of the legacy `**Plan**:` checklist. - Updates `README.md` "What it does" step 12 to mention the new flow. - Regenerates all 12 agent variants via `npx tasks generate-commands` so `commands-drift` CI stays green. Verification: `npm run build`, `npm test` (238 tests across 6 packages, all green), `npm run lint`, `npx -y @tasks-md/lint TASKS.md` all green.
fyodoriv
added a commit
that referenced
this pull request
May 21, 2026
* docs(plans): add plan-first workflow plan + examples/plan-template.md
Establishes the `docs/plans/` convention and ships the canonical template
that the new `/next-task` "Plan and validate" step references. The plan
file at `docs/plans/next-task-plan-first-workflow.md` documents the
workflow change itself (meta-bootstrap) and was validated by a reviewer
subagent in two passes (first pass: needs-revision with 6 concerns;
second pass: approved after the revisions addressed each concern).
Why: the existing "Plan (complex tasks only)" rule lets the same agent
both write and execute the plan — no second opinion. Plans-in-TASKS.md
are write-only. A separate plan file + reviewer-subagent validation
makes deliberation external, diffable, and re-runnable, while preserving
the "trivial fixes skip planning" exemption.
Acceptance criterion 8 of the plan (gate: `**Verdict**: approved` must
appear in the plan file before any code-commit lands) is satisfied:
awk '/^## Reviewer verdict\$/,0' docs/plans/next-task-plan-first-workflow.md \
| grep '^- \*\*Verdict\*\*:' | tail -1
→ - **Verdict**: approved
* feat(commands): add plan-and-validate workflow to /next-task
Replaces the "Plan (complex tasks only)" section in `commands/next-task.md`
with a 5-step "Plan and validate" workflow:
1. Check for an existing validated plan at `docs/plans/<task-id>.md`
2. Write the plan from `examples/plan-template.md`
3. Validate with a reviewer subagent (`reviewer` profile, fallback chain
`qa-engineer` → `researcher`); the subagent appends a `## Reviewer
verdict` block — only `approved` lets work proceed; `needs-revision`
triggers up to 3 revision cycles; `reject` halts and reports
4. Pre-register the rule-#9 metric in the task block
5. Implement
Trivial tasks (single file, <30 minutes, obvious fix) keep skipping the
plan step. The legacy `**Plan**:` inline-checklist field in TASKS.md is
superseded for the canonical workflow; the linter still accepts it for
backwards compat.
Why: today the same agent both writes and executes the plan. No second
opinion. A separate plan file + reviewer-subagent validation makes
deliberation external, diffable, and re-runnable. See the plan in
`docs/plans/next-task-plan-first-workflow.md` for the full motivation
and risk analysis.
This commit also:
- Updates `commands/lint-tasks.md` with a prose-only convention paragraph
(no linter binary change; hard CI gate deferred to follow-up task
`lint-tasks-md-plan-required-gate`).
- Updates `AGENTS.md` "Task Queue Policy" section to point at the new
workflow instead of the legacy `**Plan**:` checklist.
- Updates `README.md` "What it does" step 12 to mention the new flow.
- Regenerates all 12 agent variants via `npx tasks generate-commands`
so `commands-drift` CI stays green.
Verification: `npm run build`, `npm test` (238 tests across 6 packages,
all green), `npm run lint`, `npx -y @tasks-md/lint TASKS.md` all green.
7 tasks
fyodoriv
added a commit
that referenced
this pull request
May 21, 2026
Lift the rule-#9 pre-registration check (Hypothesis, Success/Acceptance, Pivot, Measurement, Anchor) from Minsky's repo-local scripts/check-rule-9-tasksmd-fields.mjs into @tasks-md/lint as a first-class opt-in rule. The pattern prevents post-hoc fishing for flattering metrics (Munafò et al. 2017) and pre-registers the give-up criterion (Ries 2011, pivot-or-persevere) by forcing the team to declare what observable a non-trivial change is expected to move *before* the code is written. spec.md § rule-9-pre-registration-block already documents the fields (PR #78); this PR ships the deterministic gate. Why it's opt-in: - Most TASKS.md files won't want every task pre-registered. Teams that want machine-enforced rule-#9 discipline pass `--require-prereg`. - Adopting on an existing repo is a two-step ratchet: `--prereg-allowlist=<file>` grandfathers legacy IDs while preventing *new* tasks from regressing. Allowlisted blocks still surface in the count so the operator sees remaining backfill debt. Surface: - `tasks-lint --require-prereg TASKS.md` — enforce the five fields - `tasks-lint --require-prereg --prereg-allowlist=.prereg-allowlist TASKS.md` — ratchet adoption - `import { parseRule9Blocks, classifyRule9Blocks, parseAllowlistFile, lintRule9Content } from "@tasks-md/lint"` — programmatic API Tests: 16 new tests (62 total in @tasks-md/lint); parser/MCP unaffected. --- _🤖 Written by an agent, not Fyodor. Ping me if this looks off._
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.
Summary
find ~/appscross-repo search from all next-task command variantscommands/next-task.md) and regenerated all 6 agent variantsTest plan
grep -r "find ~/apps" commands/ README.mdreturns no resultsGenerated with Devin