Skip to content

fix(opencode): use instructions array, not agent.build.prompt #60

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
chubes4 merged 1 commit into from
Apr 22, 2026
Merged

fix(opencode): use instructions array, not agent.build.prompt #60

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 25 additions & 15 deletions runtimes/opencode.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,12 +172,6 @@ wp-content/uploads/datamachine-files/agents/${DM_DRY_SLUG}/MEMORY.md
wp-content/uploads/datamachine-files/users/USER_ID/USER.md"
log "Dry-run: using placeholder agent paths (slug: $DM_DRY_SLUG)"
fi

# Build prompt from discovered files
OPENCODE_PROMPT="{file:./AGENTS.md}"
while IFS= read -r rel_path; do
OPENCODE_PROMPT="$OPENCODE_PROMPT\\\\n{file:./${rel_path}}"
done <<< "$DM_AGENT_FILES"
}

runtime_generate_config() {
Expand Down Expand Up @@ -237,15 +231,31 @@ runtime_generate_config() {
OPENCODE_JSON="$OPENCODE_JSON,\n \"plugin\": [${OPENCODE_PLUGINS}\n ]"
fi

# Agent prompt config
OPENCODE_JSON="$OPENCODE_JSON,\n \"agent\": {"
OPENCODE_JSON="$OPENCODE_JSON\n \"build\": {"
OPENCODE_JSON="$OPENCODE_JSON\n \"prompt\": \"${OPENCODE_PROMPT}\""
OPENCODE_JSON="$OPENCODE_JSON\n },"
OPENCODE_JSON="$OPENCODE_JSON\n \"plan\": {"
OPENCODE_JSON="$OPENCODE_JSON\n \"prompt\": \"${OPENCODE_PROMPT}\""
OPENCODE_JSON="$OPENCODE_JSON\n }"
OPENCODE_JSON="$OPENCODE_JSON\n }"
# Memory files as top-level "instructions" array.
#
# Why not "agent.build.prompt"/"agent.plan.prompt": setting those overrides
# OpenCode's canonical system prompt opening (<environment>...Skills provide...),
# which puts our memory at the top of system[1]. Anthropic's third-party-app
# detector fingerprints the first bytes of system[1] for Claude Max OAuth;
# when the canonical opening isn't there, it routes the request through
# extra-usage billing and returns HTTP 400:
# "Third-party apps now draw from your extra usage..."
#
# "instructions" loads each file via OpenCode's native Instruction.system()
# (packages/opencode/src/session/instruction.ts), which appends them with
# the "Instructions from: <path>" prefix AFTER the canonical opening — same
# mechanism that auto-discovers AGENTS.md. Keeps the OAuth flow intact.
if [ -n "$DM_AGENT_FILES" ]; then
OPENCODE_INSTRUCTIONS=""
while IFS= read -r rel_path; do
[ -z "$rel_path" ] && continue
OPENCODE_INSTRUCTIONS="${OPENCODE_INSTRUCTIONS}\n \"./${rel_path}\","
done <<< "$DM_AGENT_FILES"
if [ -n "$OPENCODE_INSTRUCTIONS" ]; then
OPENCODE_INSTRUCTIONS=$(echo "$OPENCODE_INSTRUCTIONS" | sed 's/,$//')
OPENCODE_JSON="$OPENCODE_JSON,\n \"instructions\": [${OPENCODE_INSTRUCTIONS}\n ]"
fi
fi

# Permission: allow DM workspace as external directory
OPENCODE_JSON="$OPENCODE_JSON,\n \"permission\": {"
Expand Down
Loading