| S1 |
Symbol Index |
PARTIAL |
ctags integration was already present in Agent 2 pre-existing content. Not added to docs but was never missing from the pipeline. |
| S2 |
TDD Test-First Loop |
PARTIAL |
Test-first loop already existed in Agent 3 Per-Task Cycle. Content was pre-existing; not newly introduced but documented in the Per-Task Execution Cycle on builder.html. |
| S3 |
Parallel Task Execution |
DONE |
parallel_runner.go (Go goroutine batch executor) + Agent 3 parallel batch protocol with write-conflict detection. See Pipeline Scripts. |
| S4 |
Staged Write Overlay |
DONE |
staged_write.py with colored terminal diff and Apply / Edit / Reject prompt. Every file write passes through this gate. See Builder â Pipeline Script Integration. |
| S5 |
Auto-Commit + PR Body |
DONE |
commit_by_task.py groups WRITE_LOG.jsonl by task and creates one conventional git commit per task. Generates PR_BODY.md. Invoked by Verifier post-SHIP hook. |
| S6 |
Review Rubric |
DONE |
templates/.agent/REVIEW_RUBRIC.md â 10-dimension Verifier scoring rubric. Loaded by Agent 6 at activation. |
| S7 |
Done Criteria Checklist |
DONE |
templates/.agent/DONE_CRITERIA.md â Builder activation checklist. Agent 3 reads this at activation to confirm readiness before beginning task execution. |
| S8 |
Hot Compilation Check |
DONE |
Behavioral rule in Agent 3 step 10c: run go build / npx tsc --noEmit / cargo check immediately after each file write. Compilation failure re-enters implementation without proceeding to tests. |
| S9 |
Environment Bootstrap |
DONE |
bootstrap.py validates toolchain, installs deps, generates .env.example. Invoked at Agent 3 activation step before any build tasks. See Pipeline Scripts. |
| S10 |
Multi-Modal Discovery |
DONE |
Image analysis protocol in Agent 1: native Claude vision processes wireframes, mockups, and screenshots attached during Discovery. No external dependency â Claude handles natively. |
| S11 |
Spec Formal Validation |
DONE |
spec_validate.py runs before the !build gate. Checks task_type, output_files, TC status, circular dependencies, and parallelism map coverage. Exit 1 blocks the gate with specific errors. |
| S12 |
Model-Adaptive Prompting |
DONE |
Behavioral adaptation tables in Agents 1â3 and 5â6. Each agent adjusts prompt structure, verbosity, and tool use for Claude, Kimi, GPT-4o, and Ollama-hosted models. |
| S13 |
Pipeline Replay |
DONE |
pipeline_replay.py restores the codebase to any task checkpoint and marks later tasks PENDING. Invoked by !replay TASK-NNN in the Orchestrator. See Fork & Replay Commands. |
| S14 |
Continuous SAST |
DONE |
sast_scan.py runs non-blocking after every approved write. Supports gosec, bandit, semgrep with inline fallback. Findings in SAST_FINDINGS.jsonl are consumed by Agent 5 (Security). See Agent 3 step 12d. |
| S15 |
Session Forking |
DONE |
session_fork.py forks the pipeline into two parallel branches. Sub-commands: create, compare, merge, status. Invoked by !fork and !merge commands in the Orchestrator. |
| S16 |
Container Verification |
DONE |
container_verify.py builds the Docker image and runs the full test suite inside the container. Auto-generates Dockerfile.test. Invoked by Verifier Phase 12B. See Verifier. |
| S17 |
Smart Retry Strategies |
DONE |
Failure classification system in Agent 3 Repair Mode with 6 failure modes: compile error, test failure, contract violation, coverage below threshold, secret found, and write rejected. Each mode has a distinct repair path. |
| S18 |
Self-Improving Pipeline |
DONE |
pipeline_improve.py analyzes post-run telemetry and generates PROP-NNN improvement proposals. Promoted proposals modify the agents directory. Invoked by !pipeline-health and !pipeline-promote. |
| S19 |
CI/CD Generation |
DONE |
Agent 2 standard CICD task rules: every project plan includes tasks for ci.yml, deploy.yml, security.yml, and goreleaser config. Generated as part of Phase B contract output. |
| S20 |
Observability Scaffolding |
DONE |
Agent 2 standard OBS task rules: every project plan includes structured logger, metrics endpoint, health check, and distributed tracing scaffold tasks automatically injected into TASK-GRAPH.md. |
| S21 |
Natural Language Gate Hints |
DONE |
Behavioral rule in Orchestrator: when a human types intent phrases ("I think we're ready", "looks good", "what's next"), the Orchestrator recognizes the intent and suggests the appropriate gate command rather than silently waiting. |
| S22 |
Secret Scanning |
DONE |
secret_scan.py with 14 regex patterns. Runs as a hard block before every file write in Agent 3. Write is refused and repair context returned if any secret pattern matches. See Builder â Pipeline Script Integration. |
| S23 |
Code Coverage Enforcement |
DONE |
check_coverage.py enforces 70% threshold across go/node/python/rust stacks. Runs in Agent 3 step 11b after TC tests pass. Also checked in Verifier Phase 3A. Below-threshold blocks DONE marking. |
| S24 |
Semantic RAG Search |
DONE |
semantic_search.py provides embedding-based codebase search. Agent 3 queries before each architecture-class task. Uses Ollama nomic-embed-text with TF-IDF fallback. See Pipeline Scripts. |
| S25 |
Database Migration Safety |
DONE |
migration_safety.py detects schema deltas, generates bidirectional SQL, and classifies data-loss risk. HIGH/DATA-LOSS risks block the pipeline until !migration-approve. See Migration Sub-Commands. |
| S26 |
Runtime API Contract Testing |
DONE |
contract_test.py starts the application and calls every endpoint in INTERFACES.md, validating status codes and response fields. Invoked by Verifier Phase 6D for runtime conformance beyond static type checking. |
| S27 |
Harness Activation Guard |
DONE |
Zero-bypass rule in CLAUDE.md: pipeline refuses all project work until Start agent N is issued as a standalone message. Gate commands equally strict â no embedded commands accepted. Pre-existing from initial design. |
| S28 |
Token Counter |
DONE |
token_tracker.py tracks session token usage per agent. Reads pricing from config/pricing.json. !budget shows total tokens, context window %, cost, and per-agent breakdown. See Pipeline Intelligence Commands. |
| S29 |
Thinking Indicators |
DONE |
âŗ status prefix rules added to all 7 agents. Each agent prefixes long-running steps with âŗ and updates to â
or â on completion. Native Claude Code spinner handles TUI animation â no custom implementation needed. |