fix(kickoff/swarm): flow-integrity and enforcement defect cluster (gh#19, gh#56, gh#57, gh#58, gh#60) - #65
Merged
dollspace-gay merged 7 commits intoAug 2, 2026
Conversation
Both dry-run guards executed after worktree creation: plan() had already
created the worktree/branch, written .kickoff-slug and PLAN_KICKOFF.md,
and recorded a permanent PlanRecord{stage:"planning"}; run() had created
its worktree and written .kickoff-slug/.kickoff-metadata.json. Repeated
dry-runs accumulated orphan worktrees and phantom "planning" rows that
cleanup classifies Stale and skips, and that pipeline reconciliation
(runs-only) never reclaims.
The guards now run before any creation. The dry-run output prints the
would-be worktree/branch (the exact names create_worktree derives) and
the prompt, unchanged in format. run --dry-run still creates the tracker
issue when none is passed - the printed prompt embeds its id.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The timeout wrapper killed the agent but nothing updated .kickoff-status, so killed agents kept RUNNING forever at the sentinel layer and only the wall-clock check against .kickoff-metadata.json ever noticed. Both launch paths (local/sandbox via build_agent_command, and the container's bash -c command) now append an exit-code-124 trailer that writes TIMEOUT to the sentinel, and normalize_status classifies TIMEOUT/timed* as "timed-out". Tests: exact-command asserts updated for the trailer; test_normalize_status_timeout_sentinel pins the classification. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…(gh#57) The swarm guide documents H2 Phase/Layer headers as the explicit-phasing form, but the design-doc parser recognized only H3 '### Phase N:' headers inside '## Requirements' - the documented form fell into unknown_sections and phasing intent silently degraded to auto-decomposition. H2 sections whose title starts with Phase/Layer (colon or space form) now parse into RequirementGroups via the same parse_layer_header/list-collapsing path, extend the flat requirements list for backward compat, and stay out of unknown_sections. Titles like "Phased Rollout" do not false-positive. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The skill's Pipeline State Initialization step wrote pipeline.json via an unconditional heredoc on every invocation - including --continue on a design whose pipeline had advanced - wiping plans/runs and regressing stage to "designed". The step now creates the file only when absent; when it exists, a jq update rewrites doc_hash (and design_doc) in place, preserving stage/plans/runs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The shipped agent_overrides (and work-check.py's built-in agent defaults) blocked destructive operations only - plain git merge, rebase, cherry-pick, reset, stash/tag/patch, and branch surgery were allowed at the hook layer, and no command was gated - while container-agents.qmd, kickoff.qmd, and the generated KICKOFF.md prompt all claimed "no push, no merge, gated commits". Enforcement now matches the claims where they were unambiguous: those commands are mechanically blocked for agents and git commit is gated on an active issue (explicit overrides still win). Plain git push remains allowed because the CI-verify flow instructs the agent to push a draft PR - blocking it would break that flow - so the prompt's Blocked Actions section and both guides now state the real contract: force-push always blocked, plain push reserved for the CI-verification steps. Includes the CHANGELOG entries for the gh#19/60/57/56/58 cluster. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
12 tasks
test_kickoff_dry_run_creates_kickoff_md pinned the pre-gh#19 behavior: it asserted the dry run CREATED the worktree and KICKOFF.md - the exact side effect gh#19 removes - and so failed CI on this branch (the local gate only ran the bin suite, missing tests/). Renamed to test_kickoff_dry_run_is_side_effect_free_and_prints_prompt: it now asserts the would-be worktree does NOT exist and checks the same prompt contract markers against stdout, where the dry run prints the prompt. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
Author
|
Review note from the vsdd-cli consumer (we verified this PR's gh#58 reconciliation against our own git-posture ruling — the PR-boundary model it ratifies is exactly right, and we've aligned our hook config to match): one residual doc overclaim survives the fix — tracking-modes.qmd:67-68 still reads "These are human-only operations. Agents should never push code", which now contradicts the PR's own deliberate plain-push carve-out ("push is reserved for the CI-verify draft-PR flow"). One sentence there would finish the docs-vs-hooks alignment this PR otherwise completes. |
Member
|
merge conflict |
…Projects#64) into fix/56-57-58-60-19 Resolves the overlap between this cluster and the merged container fixes. - launch.rs: build_agent_command auto-merged cleanly (permission_flag() refactor from Corvidae-Coding-Projects#63 + the gh#60 TIMEOUT trailer from this branch, in non-overlapping regions); launch_container's command resolved to carry BOTH the {skip_flag} (gh#59) and the exit-124 TIMEOUT trailer (gh#60). - kickoff/tests.rs: kept both appended tests (test_permission_flag_postures from Corvidae-Coding-Projects#63, test_normalize_status_timeout_sentinel from this branch). - CHANGELOG.md: repaired an auto-merge that silently dropped the gh#53 entry's opening line; full Fixed section audited. Full suite green (bin 2857, integration 194, smoke 159); clippy/fmt clean. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dollspace-gay
merged commit Aug 2, 2026
a153c3d
into
Corvidae-Coding-Projects:develop
6 checks passed
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
2 participants
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.
Fixes #19. Fixes #56. Fixes #57. Fixes #58. Fixes #60.
The kickoff/swarm flow-integrity and enforcement defect cluster from the vsdd kickoff bundle. One commit per issue.
gh#19 —
plan/run --dry-runside effects (2cf4fbd4)Both dry-run guards ran after worktree creation (
planalso after its permanentPlanRecordwrite), so repeated dry-runs accumulated orphan worktrees and phantom "planning" rows no cleanup flag reclaims. The guards now run before any creation and print the would-be worktree/branch/agent names. One deliberate remainder:run --dry-runstill creates the tracker issue when none is passed, since the printed prompt embeds its id.gh#60 — TIMEOUT sentinel never written (
d6ee70c3)Both launch paths (local/sandbox and container) now append an exit-code-124 trailer that writes
TIMEOUTto.kickoff-statuswhen the timeout wrapper kills the agent;normalize_statusclassifies it astimed-out. Killed agents previously keptRUNNINGat the sentinel layer forever.gh#57 — documented H2 phasing headers dropped (
a4d80b40)H2
## Phase:/## Layer:sections (the form the swarm guide documents) now parse intoRequirementGroups via the same header/list machinery as the H3-under-Requirements form, extend flat requirements for backward compat, and no longer fall intounknown_sections. "Phased Rollout"-style titles don't false-positive.gh#56 —
/design --continuewipes pipeline.json (a977e8a2)The skill's pipeline-state step now creates the file only when absent; when it exists it updates
doc_hash/design_docin place via jq, preservingstage/plans/runs.gh#58 — hook enforcement vs documented contract (
06ac4a67)Agent hook enforcement (
agent_overrides+ work-check.py agent defaults) now mechanically blocksgit merge/rebase/cherry-pick/reset, stash/tag/patch, and branch surgery, and gatesgit commiton an active issue — matching the claims in the guides and generated prompt. Plaingit pushstays allowed deliberately: the CI-verify flow instructs the agent to push a draft PR, so blocking it would break that flow. The prompt's Blocked Actions section and both guides now state that contract accurately (force-push always blocked; plain push reserved for the CI-verification steps) instead of overclaiming "no push".Testing
test_parse_h2_phase_headers_documented_form,test_parse_h2_layer_numbered_and_no_false_positive,test_normalize_status_timeout_sentinel; exact-command asserts updated for the TIMEOUT trailer.py_compileclean on work-check.py; full bin suite green.-D warnings -W clippy::unwrap_used -W clippy::expect_used) andcargo fmt --checkclean.CHANGELOG entries are placed mid-
### Fixed(after the merged gh#48 entry) so this PR, #63, and #64 merge cleanly in any order — verified pairwise with 3-way merge simulations.Part of the vsdd kickoff-bundle epic (magnificentlycursed#2).
Authored by Claude Code on behalf of @magnificentlycursed.
Generated with Claude Code