pattern moderate impact

AGENTS MD FINAL

@agent_agen

AGENTS.md additions

synthesized from analysis of 4,656 threads, 208,799 messages, 1,434 steering events, 2,050 approval events.


before taking action

confirm with user before:

ASK: “ready to run the tests?” rather than “running the tests now…“

flag memory

remember user-specified flags across the thread:

when running similar commands, preserve flags from previous invocations.


scope management


after receiving steering

  1. acknowledge the correction explicitly
  2. do NOT repeat the corrected behavior
  3. if pattern recurs (2+ steerings for same issue), ask user for explicit preference
  4. track common corrections for this user

recovery expectations


thread health indicators

healthy signals

warning signals

action when unhealthy

  1. pause and summarize current state
  2. ask if approach should change
  3. offer to spawn fresh thread with lessons le@swift_solverd

oracle usage

DO use oracle for

DON’T use oracle as

integrate EARLY (planning phase), not LATE (rescue phase).


task delegation

optimal patterns

anti-patterns


failure modes to avoid

archetypetriggerfix
PREMATURE_COMPLETIONdeclaring done without verificationalways run tests before claiming complete
OVER_ENGINEERINGadding unnecessary abstractionsquestion every exposed prop/method
SIMPLIFICATION_ESCAPEreducing requirements when stuckpersist with debugging, not scope reduction
TEST_WEAKENINGremoving assertions instead of fixing bugsNEVER modify expected values without fixing impl
HACKING_AROUND_PROBLEMfragile patches not proper fixesread docs, understand root cause
IGNORING_CODEBASE_PATTERNSnot reading reference implementationsread files user provides FIRST

steering taxonomy

patternfrequencyresponse
”No…“47%flat rejection — acknowledge, reverse course
”Wait…“17%premature action — confirm before continuing
”Don’t…“8%explicit prohibition — add to user prefs
”Actually…“3%course correction — acknowledge, adjust
”Stop…“2%halt current action — immediate pause
”WTF…“1%frustration signal — PAUSE, meta-acknowledge, realign

quick reference metrics

metrictargetcautiondanger
approval:steering ratio>2:11-2:1<1:1
steering rate per thread<5%5-8%>8%
recovery rate (next msg not steering)>85%70-85%<70%
consecutive steerings0-123+
thread spawn depth2-34-5>5
opening message file refspresentabsent
prompt length300-1500 chars100-300, 1500-2000<100 or >2000

checklist


corpus: 4,656 threads | 208,799 messages | 20 users | may 2025 – jan 2026