fix(transcript): prefix truncation marker

[devinoldenburg]

Prefix the truncation marker so long transcripts satisfy the documented contract and the regression test.\n\nValidation:

opencode-goal-mode@0.9.4 test
node --test

▶ resolveConfig
✔ applies env over defaults (0.46525ms)
✔ plugin options win over env (0.081292ms)
✔ keeps defaults when empty options (0.074ms)
✔ resolveConfig (1.400541ms)
▶ parseModelRef
✔ parses provider/model (0.465875ms)
✔ returns null for invalid (0.0685ms)
✔ parseModelRef (0.622792ms)
▶ parseEvaluatorResponse
✔ parses YES with reason (0.463708ms)
✔ parses NO with reason (0.182ms)
✔ rejects garbage (0.088042ms)
✔ parseEvaluatorResponse (1.626959ms)
▶ buildEvaluatorPrompt
✔ includes the completion condition in the prompt (0.112041ms)
✔ includes the transcript in the prompt (0.045292ms)
✔ handles empty transcript gracefully (0.039125ms)
✔ handles null-ish transcript gracefully (0.053625ms)
✔ asks for YES or NO format (0.069791ms)
✔ handles very long conditions without truncating the format instructions (0.077791ms)
✔ handles conditions with special characters (0.05975ms)
✔ produces deterministic output for the same inputs (0.04425ms)
✔ buildEvaluatorPrompt (0.660834ms)
▶ install.mjs
✔ dry-run and install into temp target (63.122208ms)
✔ install.mjs (63.546125ms)
▶ decideAfterIdle
✔ no-ops without condition (0.4535ms)
✔ no-ops when not on goal agent (0.302958ms)
✔ achieves on evaluator YES (0.927667ms)
✔ continues on evaluator NO (0.130417ms)
✔ pauses on no-progress streak (0.071708ms)
✔ honours abort suppress window (0.071166ms)
✔ exports consecutive cap constant (0.043541ms)
✔ decideAfterIdle (3.040917ms)
▶ goal_set gates idle loop
✔ goal_set sets condition (0.900583ms)
✔ chat.params tracks goal agent (0.370167ms)
✔ goal_set gates idle loop (1.620875ms)
✔ synthetic goal turns keep the session on the Goal agent (0.152417ms)
▶ buildAchievedSummaryBody
✔ asks for Goal Completed summary without re-verification (0.57525ms)
✔ buildAchievedSummaryBody (1.125792ms)
▶ buildContinuationBody
✔ discourages completion summary on continue (0.624083ms)
✔ buildContinuationBody (0.706542ms)
▶ resolveSessionApi
✔ uses v2 flat params for promptAsync on the injected SDK client (0.652417ms)
✔ falls back to v1 { path, body } shape when flat params throw (0.160292ms)
✔ uses HTTP when SDK client is unavailable and serverUrl is set (12.204291ms)
✔ resolveSessionApi (13.71325ms)
▶ formatTranscript
✔ returns empty string for empty messages (0.790916ms)
✔ returns empty string for messages with no text parts (0.093209ms)
✔ labels user and assistant messages (0.274333ms)
✔ labels agent messages as assistant (0.113541ms)
✔ skips synthetic parts (0.109584ms)
✔ handles non-text parts gracefully (0.068458ms)
✔ handles parts with content field instead of text (0.0625ms)
✔ detects role from m.role when info.role is absent (0.049917ms)
✔ defaults to user when role and agent are both absent (0.081083ms)
✔ applies truncation when transcript is very large (0.973416ms)
✔ does not truncate short transcripts (0.104333ms)
✔ handles messages from info.parts when m.parts is absent (0.081459ms)
✔ formatTranscript (3.420667ms)
▶ lastAssistantText
✔ returns last assistant text (0.107917ms)
✔ skips synthetic parts (0.043166ms)
✔ returns empty string when no assistant messages exist (0.05975ms)
✔ returns empty string for empty array (0.026209ms)
✔ handles agent-labeled messages as assistant (0.047625ms)
✔ handles multibyte characters (0.03725ms)
✔ lastAssistantText (0.41775ms)
ℹ tests 50
ℹ suites 12
ℹ pass 50
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 109.613167

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.