Using the Harness
This is a team operating harness: one shared contract for humans and AI agents
working in the same repository — same rules, same risk tiers, same evidence, same
definition of done. It exists to make a mixed human–AI team coherent, not to make one
agent faster.
A skill supplies a repeatable procedure. It never supplies permission or accountability.
First actions, every session
- Ground in the contract. If this project has adopted the operating model, read its
project profile before anything else (
docs/operating-model/PROJECT-OPERATING-PROFILE.md),
then the manual it binds. Profile facts beat inference.
- Classify risk before acting. Map the requested work to R0–R3. Round ambiguity
upward. Confirm authority at that tier before any consequential, destructive,
external, or credential-touching action.
- Route before doing. Check the routing table below. If a skill applies to the
work, use it. If you are unsure which applies, use
delivery-orchestrator.
Core invariants — not negotiable
- Classify risk before confirming authority. Approval is scoped to actual risk.
- Infer intent; never infer permission. Useful intent is not a grant of authority.
- Receipts, not polish. Produce an artefact or a pass/fail verdict, not a narrative.
- Review binds to the exact candidate. Any change after review invalidates it.
seed is not active. A seed profile supports design and R0/R1 work only.- Never invent a project fact. Unknowns get an owner and a resolving trigger.
- Insufficient evidence is a stop, not a prompt to improvise. Tag the roster role
that owns the decision; silence never converts to permission.
Skill routing
| Work in front of you |
Skill |
| New repo, or operating model install/repair |
operating-model-bootstrap |
| Multi-track epic, or "which skill applies?" |
delivery-orchestrator |
| Significant technical decision or ADR |
the-architect |
| New feature, epic, or change — plan before code |
spec-first-delivery |
| High-stakes decision, or before claiming done |
adversarial-gate |
| Agent output feeding a decision, store, or another agent |
domain-validator |
| Any PR or agent-generated code before commit |
pr-reviewer |
| Anything about to ship to a real environment |
release-readiness |
| CI triggers, Actions billing, issues, labels, branch protection, or release-tag workflow |
github-manager |
| Release process governance: SemVer, ADR, changelog, release authority |
release-manager |
| Enterprise policy, compliance, governance, or approved-vendor alignment |
governance-guardrail |
| Status, standup, or situation report |
sitrep |
| Plugin-directory, marketplace, or curated-list submission |
plugin-submission |
| LLM/cloud cost estimate or architecture with LLM calls |
cost-guardrail |
| Google ADK multi-agent design |
adk-expert |
| Any cloud-vendor workload |
gcp-expert / aws-expert / azure-expert / alibaba-expert (the-architect routes) |
| Adding or reviewing an MCP server |
mcp-server-scaffold |
Companion plugins
Agent-craft capabilities may be provided by installed companion plugins instead of these
skills — see INTEGRATIONS.md for the evaluated map (simplicity discipline, TDD
methodology, frontend design review, and similar craft lanes).
- Route craft work to an installed companion when the project profile names one as the
provider. Do not re-implement a capability an installed companion already provides.
- Companions supply capability, never authority. This harness's risk tiers, evidence,
and review gates still apply to their output.
- No companion installed? The baseline skills above are the fallback. Record the
equivalent and its owner in the project profile; do not silently skip the capability.
Harness tool mapping
Skills name actions, not tools. The per-harness translation lives outside the
canonical skill bodies:
- Antigravity (Gemini) — read
references/antigravity-tools.md (in this skill's
directory) before acting; it maps subagent dispatch, task tracking, and file/search
actions to Antigravity's native tools.
- Kimi Code — the plugin manifest's
skillInstructions carry the mapping.
- Claude Code / Codex — the skills' action vocabulary resolves to the native
tools of the same names.
Anti-rationalization table
| Excuse the agent makes |
Counter |
| "The skills are just docs — I can skip them" |
The skills are the harness. Skipping them is running without the contract. |
| "This task is too small for a skill" |
Gates are proportional to risk — R0/R1 is cheap. Skipped small gates are where big failures rehearse. |
| "The companion plugin handles quality" |
Companions own craft. Authority, evidence, and completion semantics stay here. |
| "I'll edit the skill body to fit my tools" |
Porting adds a thin adapter or tool mapping. Canonical skill bodies never change per harness. |
| "I remember the contract from earlier" |
Sessions compact and clear. Re-ground in the tracked profile; memory is not the contract. |
| "The user seems in a hurry" |
Speed is easy. Safe speed is engineered. Hurry raises risk tier, it does not lower it. |
When unsure
Route through delivery-orchestrator. Record every unresolved fact with an owner and
the trigger that resolves it. Stop before any R2/R3 mutation or external action that
lacks explicit authority and runnable controls.
1---2name: using-the-harness3description: Session-start orientation for this team harness. Load first, before any work. Explains the operating contract, when each skill triggers, and how installed companion plugins supply agent-craft capabilities while this harness supplies the team contract. Trigger at the start of every session and whenever unsure which skill applies.4---56# Using the Harness78This is a **team operating harness**: one shared contract for humans and AI agents9working in the same repository — same rules, same risk tiers, same evidence, same10definition of done. It exists to make a mixed human–AI team coherent, not to make one11agent faster.1213A skill supplies a repeatable procedure. It never supplies permission or accountability.1415## First actions, every session16171. **Ground in the contract.** If this project has adopted the operating model, read its18 project profile before anything else (`docs/operating-model/PROJECT-OPERATING-PROFILE.md`),19 then the manual it binds. Profile facts beat inference.202. **Classify risk before acting.** Map the requested work to R0–R3. Round ambiguity21 upward. Confirm authority at that tier before any consequential, destructive,22 external, or credential-touching action.233. **Route before doing.** Check the routing table below. If a skill applies to the24 work, use it. If you are unsure which applies, use `delivery-orchestrator`.2526## Core invariants — not negotiable2728- **Classify risk before confirming authority.** Approval is scoped to actual risk.29- **Infer intent; never infer permission.** Useful intent is not a grant of authority.30- **Receipts, not polish.** Produce an artefact or a pass/fail verdict, not a narrative.31- **Review binds to the exact candidate.** Any change after review invalidates it.32- **`seed` is not `active`.** A seed profile supports design and R0/R1 work only.33- **Never invent a project fact.** Unknowns get an owner and a resolving trigger.34- **Insufficient evidence is a stop, not a prompt to improvise.** Tag the roster role35 that owns the decision; silence never converts to permission.3637## Skill routing3839| Work in front of you | Skill |40|---|---|41| New repo, or operating model install/repair | `operating-model-bootstrap` |42| Multi-track epic, or "which skill applies?" | `delivery-orchestrator` |43| Significant technical decision or ADR | `the-architect` |44| New feature, epic, or change — plan before code | `spec-first-delivery` |45| High-stakes decision, or before claiming done | `adversarial-gate` |46| Agent output feeding a decision, store, or another agent | `domain-validator` |47| Any PR or agent-generated code before commit | `pr-reviewer` |48| Anything about to ship to a real environment | `release-readiness` |49| CI triggers, Actions billing, issues, labels, branch protection, or release-tag workflow | `github-manager` |50| Release process governance: SemVer, ADR, changelog, release authority | `release-manager` |51| Enterprise policy, compliance, governance, or approved-vendor alignment | `governance-guardrail` |52| Status, standup, or situation report | `sitrep` |53| Plugin-directory, marketplace, or curated-list submission | `plugin-submission` |54| LLM/cloud cost estimate or architecture with LLM calls | `cost-guardrail` |55| Google ADK multi-agent design | `adk-expert` |56| Any cloud-vendor workload | `gcp-expert` / `aws-expert` / `azure-expert` / `alibaba-expert` (`the-architect` routes) |57| Adding or reviewing an MCP server | `mcp-server-scaffold` |5859## Companion plugins6061Agent-craft capabilities may be provided by installed companion plugins instead of these62skills — see `INTEGRATIONS.md` for the evaluated map (simplicity discipline, TDD63methodology, frontend design review, and similar craft lanes).6465- Route craft work to an installed companion when the project profile names one as the66 provider. Do not re-implement a capability an installed companion already provides.67- Companions supply capability, never authority. This harness's risk tiers, evidence,68 and review gates still apply to their output.69- No companion installed? The baseline skills above are the fallback. Record the70 equivalent and its owner in the project profile; do not silently skip the capability.7172## Harness tool mapping7374Skills name actions, not tools. The per-harness translation lives outside the75canonical skill bodies:7677- **Antigravity (Gemini)** — read `references/antigravity-tools.md` (in this skill's78 directory) before acting; it maps subagent dispatch, task tracking, and file/search79 actions to Antigravity's native tools.80- **Kimi Code** — the plugin manifest's `skillInstructions` carry the mapping.81- **Claude Code / Codex** — the skills' action vocabulary resolves to the native82 tools of the same names.8384## Anti-rationalization table8586| Excuse the agent makes | Counter |87|---|---|88| "The skills are just docs — I can skip them" | The skills are the harness. Skipping them is running without the contract. |89| "This task is too small for a skill" | Gates are proportional to risk — R0/R1 is cheap. Skipped small gates are where big failures rehearse. |90| "The companion plugin handles quality" | Companions own craft. Authority, evidence, and completion semantics stay here. |91| "I'll edit the skill body to fit my tools" | Porting adds a thin adapter or tool mapping. Canonical skill bodies never change per harness. |92| "I remember the contract from earlier" | Sessions compact and clear. Re-ground in the tracked profile; memory is not the contract. |93| "The user seems in a hurry" | Speed is easy. Safe speed is engineered. Hurry raises risk tier, it does not lower it. |9495## When unsure9697Route through `delivery-orchestrator`. Record every unresolved fact with an owner and98the trigger that resolves it. Stop before any R2/R3 mutation or external action that99lacks explicit authority and runnable controls.
jpantsjoha