Generalise the harness and close the defects that made runs look verified - #70

Draft

KarthikAvinashFI wants to merge 45 commits into

devfrom

experiment/autonomous-harness

Commits on Aug 31, 2026

feat(harness): give the build stage a shell, a file editor and the repository
KarthikAvinashFI
committed
ac3f9a2
fix(harness): stop probe scoring unexecuted runtime tools as passing
KarthikAvinashFI
committed
cc4ec0a
feat(harness): let the build stage choose how to work from sub-skills it reads
KarthikAvinashFI
committed
df484c3
feat(harness): register postgres as a world kind and drop the unused provision skill
KarthikAvinashFI
committed
34ecc82
refactor(harness): remove the deny-by-default gate and the tools the shell replaces
KarthikAvinashFI
committed
e1992ac
feat(harness): give scenarios an open extras region and point the skill at the model
KarthikAvinashFI
committed
bbafede
docs: record what the autonomy experiment changed and what it has not proven
KarthikAvinashFI
committed
241e321
fix(harness): let a world with nothing locally executable still save
KarthikAvinashFI
committed
7194ba6
fix(harness): point the build skill at the shell it actually has
KarthikAvinashFI
committed
e98fa60
chore: stop tracking local test-suite recordings
KarthikAvinashFI
committed
3c9eab4
docs(skills): expand the build sub-skills and add a multi-actor note
KarthikAvinashFI
committed
49a4207
fix(harness): report a readiness verdict without crashing on a missing attribute
KarthikAvinashFI
committed
dac9957
feat(harness): make a world prove it answers the agent's tools before it grades
KarthikAvinashFI
committed
893a956
feat(skills): make the voice skill teach a cold model and ship its own checks
KarthikAvinashFI
committed
bff30dc
docs(skills): route every sub-skill by its own description and record the gate
KarthikAvinashFI
committed
83a5b72
refactor(skills): adopt the canonical skill layout and let evidence choose the reference
KarthikAvinashFI
committed
ef4d33a
feat(harness): resolve the call runner from what the environment declared
KarthikAvinashFI
committed
1aec061
feat(skills): build scenario generation on the six-axis framework and state the execution doctrine
KarthikAvinashFI
committed
eda7484
docs(skills): forbid printing a credential and say what one printed line costs
KarthikAvinashFI
committed
6a86cb0
docs(skills): forbid printing a credential and say what one printed line costs
KarthikAvinashFI
committed
9b3b5b7
chore(harness): bump the hosted image revision for the experiment run
KarthikAvinashFI
committed
8f9352c
fix(harness): load a written runner in isolation and fail typed when it will not load
KarthikAvinashFI
committed
6825494
fix(harness): let a silent-agent outcome keep its own diagnosis
KarthikAvinashFI
committed
0d08c85
fix(harness): verify a world by identity so a replacement is never skipped
KarthikAvinashFI
committed
1fc2487
docs(skills): put the weight in the per-type references and keep the body under the limit
KarthikAvinashFI
committed
2b18cfb
fix(scenarios): grant both writers the tools their skill catalogue names
KarthikAvinashFI
committed
4fb2458
fix(harness): restore deny-by-default so a stage's tool grant means something again
KarthikAvinashFI
committed
cae0ee1
fix(world): register row stores as a set and say what an unknown store was assumed to be
KarthikAvinashFI
committed
6087425
fix(transports): a written runner inherits the evidence its declared transport owes
KarthikAvinashFI
committed
bd945d3
test(permissions): hold the gate against qualified harness tools and correct the containment claims
KarthikAvinashFI
committed
9b1d233
fix(scheduler): keep the outcome on the receipt when the evidence gate rejects it
KarthikAvinashFI
committed
2935402
fix(scenarios): a review that never ran is not a review that approved
KarthikAvinashFI
committed
3eed053
fix(understand): put the stage that reads an agent back on a read-only grant
KarthikAvinashFI
committed
adf46b7
docs: state why no hosted run has happened, since the reason that circulated was wrong
KarthikAvinashFI
committed
7fe0abd
docs: record the live hosted run that proved the branch end to end
KarthikAvinashFI
committed
e49e19a
fix(gate): ask whether a world can call before whether it declares, and say what was proved
KarthikAvinashFI
committed
f7873a2
fix(authoring): report a declined environment as its own failure instead of a crash
KarthikAvinashFI
committed
aca3499
fix(bundle): start the agent from the command the contract records, not a filename
KarthikAvinashFI
committed
e191e52
fix(seed): emit one ordered table-level primary key and keep NOT NULL and defaults
KarthikAvinashFI
committed
e17c57c
fix(session): keep the end of a path in a displayed label, since that is what names it
KarthikAvinashFI
committed
8f35bda
fix(contract): let an agent say its endpoint speaks neither envelope, and refuse it at contract time
KarthikAvinashFI
committed
cf00e47
fix(bundle): say when no tool-based sub-goal can pass instead of grading the agent for it
KarthikAvinashFI
committed
066714e
docs(runner): name the upload and the evidence seam the example depends on
KarthikAvinashFI
committed
a00e487
docs(experiment): drop a name and two local-only paths from a public-repo file
KarthikAvinashFI
committed
a3a1ef8
chore(hosted): bump the source revision so the sandbox image rebuilds
KarthikAvinashFI
committed
87856e9