chore(release): v2.8.1 - the green-board train (PPLAN-0047, PR#750). Version bumped in BOTH release-tied files (pyproject.toml and src/aipass/__init__.py) and the CHANGELOG's top section restamped from 2026-08-31 to 2026-09-01 with the version, because that whole block is unmerged - main's last merge was PR#749 on 08-30, so the FPLAN-0460/0461 work ships as ONE section on its real merge date and becomes the GitHub Release notes. PATCH per the standing S318 ruling; PyPI has 2.8.0 so 2.8.1 is the first free number. PRE-FLIGHT ON THE RECORD: PR#750 mergeable_state clean at head 31a60fc, 20 commits shipping. README truth check NOT triggered and that is measured, not assumed - git diff origin/main..dev touches zero of README.md, docs/ or setup.sh; the agent count was verified anyway at 18 from the registry, 18 from the directory glob, and 18 in all three README claims. THE AUDIT GATE, run twice and the first run reported wrong: piping the CI script to tail handed me tail's exit code, so a FAILED run printed exit 0 - seedgo came back 99 percent on two TRINITY findings in their live local.json (an out-of-order entry, a 310-char summary against a 300 cap). Patrick pushed back that a 99 would red CI regardless of gitignore, which was the right instinct to test rather than argue: trinity_check.py returns not_applicable on a clean checkout - no AIPASS_REGISTRY.json in any parent and no sibling carrying a .trinity dir - and branch_audit drops the standard from the gating average entirely, the docstring naming why a 0 would blame the branch for the environment and a 100 would claim a measurement that never happened. Proved rather than reasoned: git archive HEAD into a registry-free directory, the exact CI script run against it, exit code captured with dollar-question-mark - all 18 branches at 100 percent including seedgo. The local finding is real and stays seedgo's to fix; it is not a merge blocker. CARRIED FORWARD: the local gate and the CI gate measure DIFFERENT standard sets - local runs TRINITY, CI does not. Benign in this direction because local is stricter; the dangerous direction is a standard that measures differently on CI, where a local green would mean nothing. That is last night's two-universes species surfacing in the audit lane instead of the test lane, and nobody has swept it. HELD OUT of this train deliberately: seedgo's test-inventory phase A (FPLAN-0468) stays uncommitted until they land a test_inventory gitignore rule - the rows artifact is 28MB and a --all commit would put it in the public repo permanently - and their own live WIP in bypass.json and README.md is theirs to land, not mine to sweep.

Author: AIOSAI

CHANGELOG.md

@@ -9,7 +9,7 @@ PyPI version — not the changelog header.
 
 ---
 
-## [2026-08-31] — one test universe: the night CI-red stopped being a norm (FPLAN-0460/0461)
+## [2026-09-01] — one test universe: the night CI-red stopped being a norm (FPLAN-0460/0461) · v2.8.1
 
 ### Fixed
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "aipass"
-version = "2.8.0"
+version = "2.8.1"
 description = "A local multi-agent framework where your AI agents keep their memory, work together, and never ask you to re-explain context"
 readme = "README.md"
 license = "MIT"

src/aipass/__init__.py

@@ -3,4 +3,4 @@
 git clone + ./setup.sh — https://github.com/AIOSAI/AIPass
 """
 
-__version__ = "2.8.0"
+__version__ = "2.8.1"