sage-dev-test

name	sage-dev-test
description	Ad hoc dev/test cycle -- runs existing tests and fixes failures iteratively
when_to_use	When you want to run tests and fix failures without a full feature workflow (no spec, no progress file, full regression by default)

Sage Dev/Test Skill

You are the Team Lead for an ad-hoc dev/test cycle. You coordinate two agents -- Developer (fixes code) and Tester (runs tests) -- until tests pass or max_cycles is hit.

Path note: The python _tools/load_agents.py command below assumes you're in the sage-feature-team source repo. From inside an installed project (a .sage/ directory exists), use python .sage/_tools/load_agents.py instead. Agent prompts spawned from the loader already have helper-script paths baked in by load_agents.py via {SAGE_TOOLS_DIR}.

This is a stateless workflow:

• No feature spec
• No progress file
• No output_dir writes
• Just iterate until green or escalate

The actual test command, test paths, and result parsing all come from the Tester's .sage/sage-tester-config.yaml instructions -- not from this skill.

---

Step 1: Parse Input

/sage-dev-test                          # Full regression (default)
/sage-dev-test test_login test_register # Targeted tests
/sage-dev-test --max-cycles 10          # Override cycle limit

Compute these once and reuse:

• targeted_tests -- list from positional args, or empty (full regression)
• test_scope -- "full regression" if no targeted tests, else "specific tests: <names>"
• max_cycles -- from --max-cycles N if given, else from config

Don't ask the user for a feature name or requirements. Use the literal label dev_test_cycle in messages.

When you send messages to agents below, write the message naturally with these literal values inlined -- not Python f-string syntax with {...} placeholders.

---

---

Step 2: Load Agent Prompts

python _tools/load_agents.py dev-test-only

Expected JSON:

{
  "success": true,
  "mode": "dev-test-only",
  "team_name": "<from sage-config.yaml team.dev_test_team_name>",
  "agent_names": ["Developer", "Tester"],
  "agents": {
    "Developer": "<fully rendered prompt>",
    "Tester":    "<fully rendered prompt>"
  },
  "sage_dir": "<path to .sage/ directory>",
  "config_summary": { "project_name": "...", "absolute_root_dir": "..." }
}

Validate: success == true, agent_names == ["Developer", "Tester"], both in agents. If sage_dir is null, the agents will get a "no project instructions configured" placeholder -- warn and continue.

---

Step 3: Create Team and Spawn Agents

TeamCreate(team_name=team_name, description="Sage dev/test cycle team")

Agent(name="Developer", prompt=agents["Developer"], team_name=team_name, subagent_type="general-purpose")
# wait 2-3s
Agent(name="Tester",    prompt=agents["Tester"],    team_name=team_name, subagent_type="general-purpose")

Both agents idle until you send a SendMessage task -- _BASE.md enforces the Task-Waiting Rule.

---

Step 4: Initial Test Run (Discovery)

Send a SendMessage to Tester whose body includes:

• @User: [Dev-Test] Run tests and report results. (Initial discovery) opener
• [Task: tester-discovery]
• Test scope: <test_scope> (literal -- "full regression" or the targeted test names)
• Reference: HANDBOOK.md

Run starting-message + completion monitoring (Step 6).

Parse Tester's completion message:

• Extract TEST_FAILURE: <test_name> | expected=<x> | actual=<y> | error=<msg> lines
• Count passed and failed

If all passed -> jump to Step 7 (Success Report). Otherwise -> Step 5 (Cycle Loop).

---

Step 5: Dev/Test Cycle Loop

cycle_count = 1
max_cycles  = (from --max-cycles or config)

WHILE cycle_count <= max_cycles:
  [1] Send Developer task with current failing test names + previous failure details
      Run starting-message + completion monitoring
  [2] Send Tester task (same scope as discovery)
      Run starting-message + completion monitoring
      Parse TEST_FAILURE lines
  [3] If all PASSED -> break (success)
      If FAILED and cycle_count < max -> cycle_count++, loop
      If FAILED and cycle_count == max -> escalate "Max cycles exceeded"

Developer message

Send to Developer, body includes:

• @User: [Dev-Test Cycle <n>/<max>] Make failing tests pass. opener
• [Task: dev-cycle-<n>] [Cycle: <n>/<max>]
• If cycle > 1: paste previous cycle's TEST_FAILURE lines verbatim
• Failing tests to fix: followed by a bulleted list of test names
• Reference: HANDBOOK.md

Tester message

Send to Tester, body includes:

• @User: [Dev-Test Cycle <n>/<max>] Run tests and report results. opener
• [Task: tester-<n>] [Cycle: <n>/<max>]
• Test scope: <test_scope> (same value as discovery)
• Reference: HANDBOOK.md

Idle = completion

When an agent sends STATUS: DONE | READY: yes (or STATUS: FAILED) and goes idle, that idle notification IS the completion signal. Read the agent's last message and route immediately.

---

Step 6: Monitoring (Starting Message + Completion)

Same as /sage-feature-team -- see that skill's "Step 8: Monitoring" section. Briefly:

Phase	Deadline	On miss
Starting message	60s	Escalate -- send shutdown_request, treat agent as dead, abort cycle
Completion	8 min (work timeout)	Escalate -- send shutdown_request, treat as deadlocked, abort cycle

No SYN/SYN-ACK/ACK handshake. No message-ID dedup. The agent sends one starting message and one completion message; you re-read story state (or test output) to route. Send shutdown_request after each completion to clean up the worker.

---

Step 7: Final Report

Success

@User: [Dev-Test] Test/Fix Cycle Complete

Status: SUCCESS
Cycles used: {cycle_count}/{max_cycles}

All tests passing. See git diff for code changes.

Escalation

@User: [Dev-Test] Workflow Blocked

Status: ESCALATION
Blocker: <starting-message timeout | Work timeout | Max cycles exceeded | Test hang>

Details: <list failing tests, last error, agent that hung>
Recommended action: <what the user should do>

---

Key Rules

DO:

• Assume dev-test-only mode (Developer + Tester only -- no ProductOwner, no TestCreator)
• Default to full regression unless /sage-dev-test <test_names> was given
• Send the first task to Tester (initial discovery), then alternate Developer -> Tester per cycle
• Trust the Tester's .sage/ instructions for the test command -- don't hardcode one here
• Treat the agent's single completion message as authoritative -- no SYN-ACK reply, no message-ID tracking, no dedup

DON'T:

• Don't ask the user for a feature name or requirements
• Don't create or read a progress file (this workflow is stateless)
• Don't preflight the test command or runner -- Tester validates that per its project instructions
• Don't make technical decisions -- escalate to User if anything is ambiguous

---

References

• HANDBOOK.md -- Full protocol (completion reporting model, escalation, Monitor)
• guides/ORCHESTRATOR_PATTERNS.md -- Reusable patterns shared with sage-feature-team
• agents/developer.md, agents/tester.md -- Agent role files
• examples/static-site-generator/.sage/sage-tester-config.yaml -- Reference Tester config

---

Token Tracking (Record)

After reporting to the user, record this skill's estimated token consumption:

python .sage/_tools/record_worker_usage.py     --feature <feature_name> --role Developer --story - --cycle 1     --inline --output-chars <approximate output chars produced>

Inline-mode entries are flagged estimated: true in _output/<feature_name>/tokens.json because we can't measure exact tokens from inside the main conversation (use /usage for the precise session total). Estimate output-chars as roughly the size of files you wrote + your final user-facing report. Failure here is non-fatal -- log and continue.