awslabs/aidlc-workflows v2.11.1-preview.20261009.1

AI-DLC Workflow 2.11.1-preview.20261009.1

on GitHub

Warning: Full Suite failed for this source. A Full Suite failure report ends these notes.

Merged commits since preview source 71e7aa5:

• fix(review): the record says who wrote the review, and the gate says it in one line (#2250)
• fix(kiro-ide): refuse shell calls naming AIDLC in the turn a terminal command ran (#2027)
• fix(codex): commands run as the agent wrote them, so the shipped git rules match and parallel builds stop asking (#2249)
• fix(swarm): a parallel Unit lands on Codex; a worker's brief is not the person's turn (#2248)
• fix(team): a Unit whose plan the person approved is no longer held at the merge (#2247)
• fix(aida): judge findings by what a person meets in a normal run; AIDA comments, never blocks a merge (#2223)
• fix(construction): parallel Units start with Plan Approval off instead of refusing the engine's own record (#2245)
• fix(construction): swarm prepare accepts the Plan Approval answer the engine itself recorded (#2240)
• fix(hooks): a parallel Unit builds in its prepared worktree and its reviewer writes the verdict (#2239)
• fix(kiro-ide): a hook switched off in Agent Hooks is said once by the next step, and a reply it did not record names the panel (#2243)
• fix(plugins): a plugin composed on Codex gets its runner skills (#2241)
• feat(plugins): let a contribution enrich a core persona with prose fragments (#1246)
• fix(codex): an agent role file keeps every backslash in its text (#2242)
• fix(gates): a check that still fails on a stage output is said to the person when the gate opens (#2226)
• fix(doctor): a Kiro hook switched off under Agent Hooks is reported, naming what it no longer protects (#2224)
• fix(kiro-ide): after a timed-out command the agent runs one short command and carries on, and removes a lock file only when doctor names it (#2236)
• fix(swarm): a Unit lands with the person's own git: hooks run, LFS and merge-driver config no longer refuse, landed changes are recorded (#2237)
• fix(kiro): the generated knowledge preflight tells a Kiro helper to read each file in parts, to its last line (#2238)
• fix(log): an interview answer needs a reply the person sent after its question was logged (#2209) (#2230)
• fix(config): a refresh after composing a plugin is no conflict on Copilot (#2235)
• fix(construction): a strict the team locked keeps the Unit review required at completion too (#2233)
• fix(hooks): the Stop hook waits where the question is for the person, under autonomous Construction too (#2205) (#2219)
• fix(release): stable release notes come from CHANGELOG.md (#2234)
• fix(kiro-ide): the asks on a redirect, chain or substitution cover AI-DLC commands only, so Always allow holds for the project's own commands (#2229)
• fix(kiro-ide): the approval-gate floor says the person has not answered yet, not that their answer went unrecorded (#2231)
• fix(review): the findings refusal names the rule, row and cell, and a verdict line repeated word for word is one line (#2204) (#2221)
• fix(cursor): a helper's git uses the person's own pager and monitor settings (#2228)
• fix(construction): the verification question says which tests it will not run, and the code summary lists tests not executed (#2225)
• fix(git): use the person's own git settings in AI-DLC's git calls (#2222)
• fix(sensor): the linter runs the project's own ESLint, whatever its major (#2202) (#2215)
• fix(spaces): a space switch writes a git-ignored copy of the method instead of rewriting tracked includes (#2188)
• fix(kiro-ide): a project command Kiro did not run is reported as declined or waiting for approval, not as an untrusted folder (#2220)
• fix(sensor): traceability joins a Unit by ID, by directory, or by name (#2217)
• fix(kiro): the agent reads every protocol module and stage file to its last line, in parts that fit Kiro's cut (#2218)
• fix(construction): a Unit completion waits for its review verdict, and the review retry refusals name the next step (#2216)
• feat(plugins): graduate adds.requires_stage into an implemented contribution surface (#1245)
• fix(presence): a gate the person has not answered yet is not a missed reply (#2191)
• fix(log): the message proof follows the presence check's own rules (#2195)

Source commit: b8d9bdc

Nightly test report

Full Suite failed for b8d9bdc3a197 in run 37993041332. The preview build does not wait for these tests.

Failed legs: live_macos (failure), live_windows (failure).

Failing tests

• t-exec-codex-compose-inflight.serial (1 failed assertions) in live-codex-1-macOS
  • t-exec-codex-compose-inflight - in-flight recompose over exec + exec resume > beat 1 stops at the gate with nothing applied; beat 2 resume-approves the SKIP flips
• t-journey-check-off-on.sdk (1 failed assertions) in live-claude-sdk-1-Windows
  • t-journey-check-off-on (sdk): a check turned off and on from the chat is done at once and said plainly > off from the chat, the new chat says so, back on from the chat

Failed jobs

• Linux / result (1): Linux / result
• macOS / claude-sdk (1): macOS / claude-sdk 2/2
• macOS / codex (1): macOS / codex 1/1
• Windows / claude-sdk (1): Windows / claude-sdk 1/2

Full logs are in the run 37993041332 full-suite-* artifacts.