docs: add status-values.md and update README status input

[adamhenson]

Summary

• New docs/status-values.md documenting how the status input works:
  • How status is determined in companion vs inline mode
  • Built-in GitHub conclusion mapping table (normalizeConclusion)
  • Custom statuses with needs_human example (conditional YAML expression)
  • Full table of all valid AgentMeter status values and which are settable
• Updated README inputs table status row to reference the new doc and mention needs_human

Test plan

• CI passes
• AgentMeter inline test fires on this PR

Made with Cursor

[github-actions]

⚡ AgentMeter

#	Workflow	Model	Status	Cost	Duration
1	Agent: Code Review	claude-sonnet-4-5	❌	$0.00	2m
2	Agent: Code Review (Codex)	gpt-5.4-mini	✅	$0.09	34s
3	AgentMeter — Inline Test	claude-sonnet-4-5	✅	$0.01	11s
4	Agent: Code Review	claude-sonnet-4-5	✅	$0.35	5m
5	Agent: Code Review (Codex)	gpt-5.4-mini	✅	$0.24	48s
Total				$2.31

All 15 runs

#	Workflow	Model	Status	Cost	Duration
1	Agent: Code Review	claude-sonnet-4-5	❌	$0.00	2m
2	Agent: Code Review (Codex)	gpt-5.4-mini	✅	$0.09	34s
3	AgentMeter — Inline Test	claude-sonnet-4-5	✅	$0.01	11s
4	Agent: Code Review	claude-sonnet-4-5	✅	$0.35	5m
5	Agent: Code Review (Codex)	gpt-5.4-mini	✅	$0.24	48s
6	Agent: Code Review	claude-sonnet-4-5	🚫	$0.00	1m
7	Agent: Code Review (Codex)	gpt-5.4-mini	❌	$0.27	58s
8	AgentMeter — Inline Test	claude-sonnet-4-5	✅	$0.01	10s
9	AgentMeter — Inline Test	claude-sonnet-4-5	✅	$0.01	12s
10	Agent: Code Review	claude-sonnet-4-5	✅	$0.32	5m
11	Agent: Code Review (Codex)	gpt-5.4-mini	✅	$0.53	2m
12	AgentMeter — Inline Test	claude-sonnet-4-5	✅	$0.01	12s
13	Agent: Code Review	claude-sonnet-4-5	✅	$0.29	4m
14	Agent: Code Review (Codex)	gpt-5.4-mini	✅	$0.17	52s
15	AgentMeter — Inline Test	claude-sonnet-4-5	✅	$0.01	11s
Total				$2.31

Token breakdown

Type	Tokens	Cost
Input	0	$0.00
Output	0	$0.00
Cache writes	0	$0.00
Cache reads	0	$0.00

Model: claude-sonnet-4-5

View in AgentMeter →

[github-actions]

• High: docs/status-values.md:11-18 and README.md:284 claim workflow_run_id makes status resolve automatically from the triggering workflow. The implementation does not do that: src/run.ts:69-83 still passes inputs.status, and src/inputs.ts:18 defaults missing status to success. If users follow the new docs and omit status, failed companion runs will be recorded as success. Please either change the code to read github.event.workflow_run.conclusion internally or correct the docs to require status: ${{ github.event.workflow_run.conclusion }}.

Otherwise the status mapping table looks consistent with normalizeConclusion().

[github-actions]

Two issues:

• High: src/run.ts and src/workflow-run.ts can produce blank ingest data when getWorkflowRun fails. The fallback path sets resolvedWorkflowName to '' and serializes missing timestamps as '', so the API/comment can end up with empty workflow names and invalid dates instead of a clean fallback or skip. Keep the companion workflow name when the run lookup fails, and avoid sending empty timestamp strings.

• Medium: src/run.ts marks token data as approximate based on baseTokens, even when explicit token inputs override the extracted values. If a caller supplies exact token counts alongside agent_output, the final payload/comment can still be labeled approximate. Derive isApproximate from the final merged source of each field, or force it to false when all token counts are explicitly provided.

Core flow otherwise looks solid.

[github-actions]

No blocking issues found in the patch.

The cache-hit-rate change in src/comment.ts is internally consistent, and the new tests in tests/comment.test.ts cover the new denominator. I couldn’t run the suite locally because the workspace is missing installed dependencies, so the only residual risk is runtime integration rather than an obvious code defect.

[github-actions]

No correctness or production issues found in action.yml. Removing author is metadata-only and does not affect runtime behavior.