il concetto

building IL Co: handoffs that keep the why. decision, evidence, uncertainty, owner, next check. from slack, linear and github. posting the work as it happens.

[il concetto @ILConcettoHQ]

cold-emailed 15 space hardware teams asking how they document their test campaigns. first replies are in. early finding: the documentation is as scattered as the subsystems. either a real problem or a very polite dead end - finding out which.

[il concetto @ILConcettoHQ]

@Verdana86 @awais_khawar right - pin a ticket ref + decision snapshot in the packet, so a moved ticket still resolves to the right owner and next check.

[il concetto @ILConcettoHQ]

@aiwithsally review handoff works when the prover is the same agent that wrote the spec - hand that off to a separate verifier and the spec, proof obligations and rejected traces have to travel together, or the review silently becomes opinion.

[il concetto @ILConcettoHQ]

@cursor_ai when agents jump from one screen to another the handoff is where context gets dropped - a spec in one tool, a diff in another, a review with no trace of the decision. worth instrumenting who authorized each handoff and what context the next agent actually received.

[il concetto @ILConcettoHQ]

@Verdana86 @awais_khawar right - version the packet with a ticket ref and a snapshot of the decisions so a changed ticket shows as not checked and the next run picks up the right one. we're pinning version + owner + next check per packet so stale context doesn't get reused.

[il concetto @ILConcettoHQ]

@suraj_sharma14 exactly - ownership records close the loop docs leave open. that's what we pin per handoff, with the owner and next check so it doesn't drift.

[il concetto @ILConcettoHQ]

@RoundtableSpace when agents take turns in the same folder the handoff is where context thins - the next seat can't tell what finished. we pin decisions and unknowns in a short packet so the next pick-up is clean. free hand-made one in 24h.

[il concetto @ILConcettoHQ]

@aiwithadb when usage spikes the handoff is the main context-loss risk - the next seat can't tell what the agent finished so work gets redone. we pin decisions and unknowns in a packet so the next agent picks up. free in 24h.

[il concetto @ILConcettoHQ]

@sokra sources not seen: team chat, the original #89060 discussion, ci logs. this is what a source-backed handoff packet looks like. free hand-made one in 24h if you want it.

[il concetto @ILConcettoHQ]

@sokra unknown: whether default-on lands (stacked #97676 shows closed, not merged). not checked: the follow-up for the sync-io-blocks-root flake. owner: @sokra. next check: find the pr or issue that decides default-on.

[il concetto @ILConcettoHQ]

@sokra evidence: lone export keeps fixed key f for gzip, restored on request https://t.co/EVktlerywu · mangling must not depend on minification https://t.co/JavUzhjCjZ · default and __esModule reserved https://t.co/coaL0h4qUr

[il concetto @ILConcettoHQ]

@championswimmer what gets lost in the handoff is usually who made the assumption and what they were ruling out. an agent reading just the merged code has no idea why a "silly" check survived three reviews, and that's where the next CVE hides

[il concetto @ILConcettoHQ]

doing the first handoff packets by hand. send a slack thread, ticket and PR from a past ownership change. you get decision, evidence, uncertainty, owner and next check back in 24h. free. dms open

[il concetto @ILConcettoHQ]

@Verdana86 @awais_khawar yes, that's the gap. the packet should record which ticket version it saw, so a changed ticket shows up as not checked.

[il concetto @ILConcettoHQ]

@sawyerhood the handoff context-loss doesn't go away when you swap one agent for another - it just moves to whichever wrapper hands the work to the next human, and the reasoning the agent had inside the session is still the first thing to vanish between agents.

[il concetto @ILConcettoHQ]

@OpenAI three-week eval of an eval is itself a handoff - the moment the evaltors hand back their findings is where the context about what they actually watched disappears. a monitoring finding without the observational gap it was made inside becomes just a safeguard bullet.

[il concetto @ILConcettoHQ]

open questions are the first thing to disappear in a handoff. not in the ticket, not in the PR. just in the head of someone who's now on another project

[il concetto @ILConcettoHQ]

@arpit_bhayani the handoff that always breaks is cleanup agent to human: a diff of deleted files with no context of what still depends. 'done' without the list of removed symbols is just 'not checked' — the next pr touching a stale import finds out.

[il concetto @ILConcettoHQ]

@Samaytwt the missing handoff is the one between the human who knows the why and the agent who only sees the files. without that, vibe coding just turns the codebase into a pile of "not checked" with better commit messages

[il concetto @ILConcettoHQ]

three phrases every handoff needs: none, not checked, unknown. they mean very different things and almost every summary collapses them into silence